drivers/md/dm-integrity.c

   1 /*
   2  * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
   3  * Copyright (C) 2016-2017 Milan Broz
   4  * Copyright (C) 2016-2017 Mikulas Patocka
   5  *
   6  * This file is released under the GPL.
   7  */
   8
   9 #include "dm-bio-record.h"
  10
  11 #include <linux/compiler.h>
  12 #include <linux/module.h>
  13 #include <linux/device-mapper.h>
  14 #include <linux/dm-io.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/sort.h>
  17 #include <linux/rbtree.h>
  18 #include <linux/delay.h>
  19 #include <linux/random.h>
  20 #include <linux/reboot.h>
  21 #include <crypto/hash.h>
  22 #include <crypto/skcipher.h>
  23 #include <linux/async_tx.h>
  24 #include <linux/dm-bufio.h>
  25
  26 #include "dm-audit.h"
  27
  28 #define DM_MSG_PREFIX "integrity"
  29
  30 #define DEFAULT_INTERLEAVE_SECTORS      32768
  31 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  32 #define DEFAULT_SECTORS_PER_BITMAP_BIT  32768
  33 #define DEFAULT_BUFFER_SECTORS          128
  34 #define DEFAULT_JOURNAL_WATERMARK       50
  35 #define DEFAULT_SYNC_MSEC               10000
  36 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
  37 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  38 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  39 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  40 #define RECALC_SECTORS                  32768
  41 #define RECALC_WRITE_SUPER              16
  42 #define BITMAP_BLOCK_SIZE               4096    /* don't change it */
  43 #define BITMAP_FLUSH_INTERVAL           (10 * HZ)
  44 #define DISCARD_FILLER                  0xf6
  45 #define SALT_SIZE                       16
  46
  47 /*
  48  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  49  * so it should not be enabled in the official kernel
  50  */
  51 //#define DEBUG_PRINT
  52 //#define INTERNAL_VERIFY
  53
  54 /*
  55  * On disk structures
  56  */
  57
  58 #define SB_MAGIC                        "integrt"
  59 #define SB_VERSION_1                    1
  60 #define SB_VERSION_2                    2
  61 #define SB_VERSION_3                    3
  62 #define SB_VERSION_4                    4
  63 #define SB_VERSION_5                    5
  64 #define SB_SECTORS                      8
  65 #define MAX_SECTORS_PER_BLOCK           8
  66
  67 struct superblock {
  68         __u8 magic[8];
  69         __u8 version;
  70         __u8 log2_interleave_sectors;
  71         __le16 integrity_tag_size;
  72         __le32 journal_sections;
  73         __le64 provided_data_sectors;   /* userspace uses this value */
  74         __le32 flags;
  75         __u8 log2_sectors_per_block;
  76         __u8 log2_blocks_per_bitmap_bit;
  77         __u8 pad[2];
  78         __le64 recalc_sector;
  79         __u8 pad2[8];
  80         __u8 salt[SALT_SIZE];
  81 };
  82
  83 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  84 #define SB_FLAG_RECALCULATING           0x2
  85 #define SB_FLAG_DIRTY_BITMAP            0x4
  86 #define SB_FLAG_FIXED_PADDING           0x8
  87 #define SB_FLAG_FIXED_HMAC              0x10
  88
  89 #define JOURNAL_ENTRY_ROUNDUP           8
  90
  91 typedef __le64 commit_id_t;
  92 #define JOURNAL_MAC_PER_SECTOR          8
  93
  94 struct journal_entry {
  95         union {
  96                 struct {
  97                         __le32 sector_lo;
  98                         __le32 sector_hi;
  99                 } s;
 100                 __le64 sector;
 101         } u;
 102         commit_id_t last_bytes[];
 103         /* __u8 tag[0]; */
 104 };
 105
 106 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
 107
 108 #if BITS_PER_LONG == 64
 109 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 110 #else
 111 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
 112 #endif
 113 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
 114 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 115 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 116 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 117 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 118
 119 #define JOURNAL_BLOCK_SECTORS           8
 120 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 121 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 122
 123 struct journal_sector {
 124         struct_group(sectors,
 125                 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 126                 __u8 mac[JOURNAL_MAC_PER_SECTOR];
 127         );
 128         commit_id_t commit_id;
 129 };
 130
 131 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 132
 133 #define METADATA_PADDING_SECTORS        8
 134
 135 #define N_COMMIT_IDS                    4
 136
 137 static unsigned char prev_commit_seq(unsigned char seq)
 138 {
 139         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 140 }
 141
 142 static unsigned char next_commit_seq(unsigned char seq)
 143 {
 144         return (seq + 1) % N_COMMIT_IDS;
 145 }
 146
 147 /*
 148  * In-memory structures
 149  */
 150
 151 struct journal_node {
 152         struct rb_node node;
 153         sector_t sector;
 154 };
 155
 156 struct alg_spec {
 157         char *alg_string;
 158         char *key_string;
 159         __u8 *key;
 160         unsigned key_size;
 161 };
 162
 163 struct dm_integrity_c {
 164         struct dm_dev *dev;
 165         struct dm_dev *meta_dev;
 166         unsigned tag_size;
 167         __s8 log2_tag_size;
 168         sector_t start;
 169         mempool_t journal_io_mempool;
 170         struct dm_io_client *io;
 171         struct dm_bufio_client *bufio;
 172         struct workqueue_struct *metadata_wq;
 173         struct superblock *sb;
 174         unsigned journal_pages;
 175         unsigned n_bitmap_blocks;
 176
 177         struct page_list *journal;
 178         struct page_list *journal_io;
 179         struct page_list *journal_xor;
 180         struct page_list *recalc_bitmap;
 181         struct page_list *may_write_bitmap;
 182         struct bitmap_block_status *bbs;
 183         unsigned bitmap_flush_interval;
 184         int synchronous_mode;
 185         struct bio_list synchronous_bios;
 186         struct delayed_work bitmap_flush_work;
 187
 188         struct crypto_skcipher *journal_crypt;
 189         struct scatterlist **journal_scatterlist;
 190         struct scatterlist **journal_io_scatterlist;
 191         struct skcipher_request **sk_requests;
 192
 193         struct crypto_shash *journal_mac;
 194
 195         struct journal_node *journal_tree;
 196         struct rb_root journal_tree_root;
 197
 198         sector_t provided_data_sectors;
 199
 200         unsigned short journal_entry_size;
 201         unsigned char journal_entries_per_sector;
 202         unsigned char journal_section_entries;
 203         unsigned short journal_section_sectors;
 204         unsigned journal_sections;
 205         unsigned journal_entries;
 206         sector_t data_device_sectors;
 207         sector_t meta_device_sectors;
 208         unsigned initial_sectors;
 209         unsigned metadata_run;
 210         __s8 log2_metadata_run;
 211         __u8 log2_buffer_sectors;
 212         __u8 sectors_per_block;
 213         __u8 log2_blocks_per_bitmap_bit;
 214
 215         unsigned char mode;
 216
 217         int failed;
 218
 219         struct crypto_shash *internal_hash;
 220
 221         struct dm_target *ti;
 222
 223         /* these variables are locked with endio_wait.lock */
 224         struct rb_root in_progress;
 225         struct list_head wait_list;
 226         wait_queue_head_t endio_wait;
 227         struct workqueue_struct *wait_wq;
 228         struct workqueue_struct *offload_wq;
 229
 230         unsigned char commit_seq;
 231         commit_id_t commit_ids[N_COMMIT_IDS];
 232
 233         unsigned committed_section;
 234         unsigned n_committed_sections;
 235
 236         unsigned uncommitted_section;
 237         unsigned n_uncommitted_sections;
 238
 239         unsigned free_section;
 240         unsigned char free_section_entry;
 241         unsigned free_sectors;
 242
 243         unsigned free_sectors_threshold;
 244
 245         struct workqueue_struct *commit_wq;
 246         struct work_struct commit_work;
 247
 248         struct workqueue_struct *writer_wq;
 249         struct work_struct writer_work;
 250
 251         struct workqueue_struct *recalc_wq;
 252         struct work_struct recalc_work;
 253         u8 *recalc_buffer;
 254         u8 *recalc_tags;
 255
 256         struct bio_list flush_bio_list;
 257
 258         unsigned long autocommit_jiffies;
 259         struct timer_list autocommit_timer;
 260         unsigned autocommit_msec;
 261
 262         wait_queue_head_t copy_to_journal_wait;
 263
 264         struct completion crypto_backoff;
 265
 266         bool journal_uptodate;
 267         bool just_formatted;
 268         bool recalculate_flag;
 269         bool reset_recalculate_flag;
 270         bool discard;
 271         bool fix_padding;
 272         bool fix_hmac;
 273         bool legacy_recalculate;
 274
 275         struct alg_spec internal_hash_alg;
 276         struct alg_spec journal_crypt_alg;
 277         struct alg_spec journal_mac_alg;
 278
 279         atomic64_t number_of_mismatches;
 280
 281         struct notifier_block reboot_notifier;
 282 };
 283
 284 struct dm_integrity_range {
 285         sector_t logical_sector;
 286         sector_t n_sectors;
 287         bool waiting;
 288         union {
 289                 struct rb_node node;
 290                 struct {
 291                         struct task_struct *task;
 292                         struct list_head wait_entry;
 293                 };
 294         };
 295 };
 296
 297 struct dm_integrity_io {
 298         struct work_struct work;
 299
 300         struct dm_integrity_c *ic;
 301         enum req_op op;
 302         bool fua;
 303
 304         struct dm_integrity_range range;
 305
 306         sector_t metadata_block;
 307         unsigned metadata_offset;
 308
 309         atomic_t in_flight;
 310         blk_status_t bi_status;
 311
 312         struct completion *completion;
 313
 314         struct dm_bio_details bio_details;
 315 };
 316
 317 struct journal_completion {
 318         struct dm_integrity_c *ic;
 319         atomic_t in_flight;
 320         struct completion comp;
 321 };
 322
 323 struct journal_io {
 324         struct dm_integrity_range range;
 325         struct journal_completion *comp;
 326 };
 327
 328 struct bitmap_block_status {
 329         struct work_struct work;
 330         struct dm_integrity_c *ic;
 331         unsigned idx;
 332         unsigned long *bitmap;
 333         struct bio_list bio_queue;
 334         spinlock_t bio_queue_lock;
 335
 336 };
 337
 338 static struct kmem_cache *journal_io_cache;
 339
 340 #define JOURNAL_IO_MEMPOOL      32
 341
 342 #ifdef DEBUG_PRINT
 343 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 344 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 345 {
 346         va_list args;
 347         va_start(args, msg);
 348         vprintk(msg, args);
 349         va_end(args);
 350         if (len)
 351                 pr_cont(":");
 352         while (len) {
 353                 pr_cont(" %02x", *bytes);
 354                 bytes++;
 355                 len--;
 356         }
 357         pr_cont("\n");
 358 }
 359 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 360 #else
 361 #define DEBUG_print(x, ...)                     do { } while (0)
 362 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 363 #endif
 364
 365 static void dm_integrity_prepare(struct request *rq)
 366 {
 367 }
 368
 369 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
 370 {
 371 }
 372
 373 /*
 374  * DM Integrity profile, protection is performed layer above (dm-crypt)
 375  */
 376 static const struct blk_integrity_profile dm_integrity_profile = {
 377         .name                   = "DM-DIF-EXT-TAG",
 378         .generate_fn            = NULL,
 379         .verify_fn              = NULL,
 380         .prepare_fn             = dm_integrity_prepare,
 381         .complete_fn            = dm_integrity_complete,
 382 };
 383
 384 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 385 static void integrity_bio_wait(struct work_struct *w);
 386 static void dm_integrity_dtr(struct dm_target *ti);
 387
 388 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 389 {
 390         if (err == -EILSEQ)
 391                 atomic64_inc(&ic->number_of_mismatches);
 392         if (!cmpxchg(&ic->failed, 0, err))
 393                 DMERR("Error on %s: %d", msg, err);
 394 }
 395
 396 static int dm_integrity_failed(struct dm_integrity_c *ic)
 397 {
 398         return READ_ONCE(ic->failed);
 399 }
 400
 401 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
 402 {
 403         if (ic->legacy_recalculate)
 404                 return false;
 405         if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
 406             ic->internal_hash_alg.key || ic->journal_mac_alg.key :
 407             ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
 408                 return true;
 409         return false;
 410 }
 411
 412 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 413                                           unsigned j, unsigned char seq)
 414 {
 415         /*
 416          * Xor the number with section and sector, so that if a piece of
 417          * journal is written at wrong place, it is detected.
 418          */
 419         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 420 }
 421
 422 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 423                                 sector_t *area, sector_t *offset)
 424 {
 425         if (!ic->meta_dev) {
 426                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 427                 *area = data_sector >> log2_interleave_sectors;
 428                 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 429         } else {
 430                 *area = 0;
 431                 *offset = data_sector;
 432         }
 433 }
 434
 435 #define sector_to_block(ic, n)                                          \
 436 do {                                                                    \
 437         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 438         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 439 } while (0)
 440
 441 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 442                                             sector_t offset, unsigned *metadata_offset)
 443 {
 444         __u64 ms;
 445         unsigned mo;
 446
 447         ms = area << ic->sb->log2_interleave_sectors;
 448         if (likely(ic->log2_metadata_run >= 0))
 449                 ms += area << ic->log2_metadata_run;
 450         else
 451                 ms += area * ic->metadata_run;
 452         ms >>= ic->log2_buffer_sectors;
 453
 454         sector_to_block(ic, offset);
 455
 456         if (likely(ic->log2_tag_size >= 0)) {
 457                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 458                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 459         } else {
 460                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 461                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 462         }
 463         *metadata_offset = mo;
 464         return ms;
 465 }
 466
 467 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 468 {
 469         sector_t result;
 470
 471         if (ic->meta_dev)
 472                 return offset;
 473
 474         result = area << ic->sb->log2_interleave_sectors;
 475         if (likely(ic->log2_metadata_run >= 0))
 476                 result += (area + 1) << ic->log2_metadata_run;
 477         else
 478                 result += (area + 1) * ic->metadata_run;
 479
 480         result += (sector_t)ic->initial_sectors + offset;
 481         result += ic->start;
 482
 483         return result;
 484 }
 485
 486 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 487 {
 488         if (unlikely(*sec_ptr >= ic->journal_sections))
 489                 *sec_ptr -= ic->journal_sections;
 490 }
 491
 492 static void sb_set_version(struct dm_integrity_c *ic)
 493 {
 494         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
 495                 ic->sb->version = SB_VERSION_5;
 496         else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
 497                 ic->sb->version = SB_VERSION_4;
 498         else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 499                 ic->sb->version = SB_VERSION_3;
 500         else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 501                 ic->sb->version = SB_VERSION_2;
 502         else
 503                 ic->sb->version = SB_VERSION_1;
 504 }
 505
 506 static int sb_mac(struct dm_integrity_c *ic, bool wr)
 507 {
 508         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 509         int r;
 510         unsigned size = crypto_shash_digestsize(ic->journal_mac);
 511
 512         if (sizeof(struct superblock) + size > 1 << SECTOR_SHIFT) {
 513                 dm_integrity_io_error(ic, "digest is too long", -EINVAL);
 514                 return -EINVAL;
 515         }
 516
 517         desc->tfm = ic->journal_mac;
 518
 519         r = crypto_shash_init(desc);
 520         if (unlikely(r < 0)) {
 521                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 522                 return r;
 523         }
 524
 525         r = crypto_shash_update(desc, (__u8 *)ic->sb, (1 << SECTOR_SHIFT) - size);
 526         if (unlikely(r < 0)) {
 527                 dm_integrity_io_error(ic, "crypto_shash_update", r);
 528                 return r;
 529         }
 530
 531         if (likely(wr)) {
 532                 r = crypto_shash_final(desc, (__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size);
 533                 if (unlikely(r < 0)) {
 534                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 535                         return r;
 536                 }
 537         } else {
 538                 __u8 result[HASH_MAX_DIGESTSIZE];
 539                 r = crypto_shash_final(desc, result);
 540                 if (unlikely(r < 0)) {
 541                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 542                         return r;
 543                 }
 544                 if (memcmp((__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size, result, size)) {
 545                         dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
 546                         dm_audit_log_target(DM_MSG_PREFIX, "mac-superblock", ic->ti, 0);
 547                         return -EILSEQ;
 548                 }
 549         }
 550
 551         return 0;
 552 }
 553
 554 static int sync_rw_sb(struct dm_integrity_c *ic, blk_opf_t opf)
 555 {
 556         struct dm_io_request io_req;
 557         struct dm_io_region io_loc;
 558         const enum req_op op = opf & REQ_OP_MASK;
 559         int r;
 560
 561         io_req.bi_opf = opf;
 562         io_req.mem.type = DM_IO_KMEM;
 563         io_req.mem.ptr.addr = ic->sb;
 564         io_req.notify.fn = NULL;
 565         io_req.client = ic->io;
 566         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 567         io_loc.sector = ic->start;
 568         io_loc.count = SB_SECTORS;
 569
 570         if (op == REQ_OP_WRITE) {
 571                 sb_set_version(ic);
 572                 if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 573                         r = sb_mac(ic, true);
 574                         if (unlikely(r))
 575                                 return r;
 576                 }
 577         }
 578
 579         r = dm_io(&io_req, 1, &io_loc, NULL);
 580         if (unlikely(r))
 581                 return r;
 582
 583         if (op == REQ_OP_READ) {
 584                 if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 585                         r = sb_mac(ic, false);
 586                         if (unlikely(r))
 587                                 return r;
 588                 }
 589         }
 590
 591         return 0;
 592 }
 593
 594 #define BITMAP_OP_TEST_ALL_SET          0
 595 #define BITMAP_OP_TEST_ALL_CLEAR        1
 596 #define BITMAP_OP_SET                   2
 597 #define BITMAP_OP_CLEAR                 3
 598
 599 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 600                             sector_t sector, sector_t n_sectors, int mode)
 601 {
 602         unsigned long bit, end_bit, this_end_bit, page, end_page;
 603         unsigned long *data;
 604
 605         if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 606                 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 607                         sector,
 608                         n_sectors,
 609                         ic->sb->log2_sectors_per_block,
 610                         ic->log2_blocks_per_bitmap_bit,
 611                         mode);
 612                 BUG();
 613         }
 614
 615         if (unlikely(!n_sectors))
 616                 return true;
 617
 618         bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 619         end_bit = (sector + n_sectors - 1) >>
 620                 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 621
 622         page = bit / (PAGE_SIZE * 8);
 623         bit %= PAGE_SIZE * 8;
 624
 625         end_page = end_bit / (PAGE_SIZE * 8);
 626         end_bit %= PAGE_SIZE * 8;
 627
 628 repeat:
 629         if (page < end_page) {
 630                 this_end_bit = PAGE_SIZE * 8 - 1;
 631         } else {
 632                 this_end_bit = end_bit;
 633         }
 634
 635         data = lowmem_page_address(bitmap[page].page);
 636
 637         if (mode == BITMAP_OP_TEST_ALL_SET) {
 638                 while (bit <= this_end_bit) {
 639                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 640                                 do {
 641                                         if (data[bit / BITS_PER_LONG] != -1)
 642                                                 return false;
 643                                         bit += BITS_PER_LONG;
 644                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 645                                 continue;
 646                         }
 647                         if (!test_bit(bit, data))
 648                                 return false;
 649                         bit++;
 650                 }
 651         } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 652                 while (bit <= this_end_bit) {
 653                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 654                                 do {
 655                                         if (data[bit / BITS_PER_LONG] != 0)
 656                                                 return false;
 657                                         bit += BITS_PER_LONG;
 658                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 659                                 continue;
 660                         }
 661                         if (test_bit(bit, data))
 662                                 return false;
 663                         bit++;
 664                 }
 665         } else if (mode == BITMAP_OP_SET) {
 666                 while (bit <= this_end_bit) {
 667                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 668                                 do {
 669                                         data[bit / BITS_PER_LONG] = -1;
 670                                         bit += BITS_PER_LONG;
 671                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 672                                 continue;
 673                         }
 674                         __set_bit(bit, data);
 675                         bit++;
 676                 }
 677         } else if (mode == BITMAP_OP_CLEAR) {
 678                 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 679                         clear_page(data);
 680                 else while (bit <= this_end_bit) {
 681                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 682                                 do {
 683                                         data[bit / BITS_PER_LONG] = 0;
 684                                         bit += BITS_PER_LONG;
 685                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 686                                 continue;
 687                         }
 688                         __clear_bit(bit, data);
 689                         bit++;
 690                 }
 691         } else {
 692                 BUG();
 693         }
 694
 695         if (unlikely(page < end_page)) {
 696                 bit = 0;
 697                 page++;
 698                 goto repeat;
 699         }
 700
 701         return true;
 702 }
 703
 704 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 705 {
 706         unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 707         unsigned i;
 708
 709         for (i = 0; i < n_bitmap_pages; i++) {
 710                 unsigned long *dst_data = lowmem_page_address(dst[i].page);
 711                 unsigned long *src_data = lowmem_page_address(src[i].page);
 712                 copy_page(dst_data, src_data);
 713         }
 714 }
 715
 716 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 717 {
 718         unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 719         unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 720
 721         BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 722         return &ic->bbs[bitmap_block];
 723 }
 724
 725 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 726                                  bool e, const char *function)
 727 {
 728 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 729         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 730
 731         if (unlikely(section >= ic->journal_sections) ||
 732             unlikely(offset >= limit)) {
 733                 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 734                        function, section, offset, ic->journal_sections, limit);
 735                 BUG();
 736         }
 737 #endif
 738 }
 739
 740 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 741                                unsigned *pl_index, unsigned *pl_offset)
 742 {
 743         unsigned sector;
 744
 745         access_journal_check(ic, section, offset, false, "page_list_location");
 746
 747         sector = section * ic->journal_section_sectors + offset;
 748
 749         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 750         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 751 }
 752
 753 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 754                                                unsigned section, unsigned offset, unsigned *n_sectors)
 755 {
 756         unsigned pl_index, pl_offset;
 757         char *va;
 758
 759         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 760
 761         if (n_sectors)
 762                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 763
 764         va = lowmem_page_address(pl[pl_index].page);
 765
 766         return (struct journal_sector *)(va + pl_offset);
 767 }
 768
 769 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 770 {
 771         return access_page_list(ic, ic->journal, section, offset, NULL);
 772 }
 773
 774 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 775 {
 776         unsigned rel_sector, offset;
 777         struct journal_sector *js;
 778
 779         access_journal_check(ic, section, n, true, "access_journal_entry");
 780
 781         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 782         offset = n / JOURNAL_BLOCK_SECTORS;
 783
 784         js = access_journal(ic, section, rel_sector);
 785         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 786 }
 787
 788 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 789 {
 790         n <<= ic->sb->log2_sectors_per_block;
 791
 792         n += JOURNAL_BLOCK_SECTORS;
 793
 794         access_journal_check(ic, section, n, false, "access_journal_data");
 795
 796         return access_journal(ic, section, n);
 797 }
 798
 799 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 800 {
 801         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 802         int r;
 803         unsigned j, size;
 804
 805         desc->tfm = ic->journal_mac;
 806
 807         r = crypto_shash_init(desc);
 808         if (unlikely(r < 0)) {
 809                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 810                 goto err;
 811         }
 812
 813         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 814                 __le64 section_le;
 815
 816                 r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
 817                 if (unlikely(r < 0)) {
 818                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 819                         goto err;
 820                 }
 821
 822                 section_le = cpu_to_le64(section);
 823                 r = crypto_shash_update(desc, (__u8 *)&section_le, sizeof section_le);
 824                 if (unlikely(r < 0)) {
 825                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 826                         goto err;
 827                 }
 828         }
 829
 830         for (j = 0; j < ic->journal_section_entries; j++) {
 831                 struct journal_entry *je = access_journal_entry(ic, section, j);
 832                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 833                 if (unlikely(r < 0)) {
 834                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 835                         goto err;
 836                 }
 837         }
 838
 839         size = crypto_shash_digestsize(ic->journal_mac);
 840
 841         if (likely(size <= JOURNAL_MAC_SIZE)) {
 842                 r = crypto_shash_final(desc, result);
 843                 if (unlikely(r < 0)) {
 844                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 845                         goto err;
 846                 }
 847                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 848         } else {
 849                 __u8 digest[HASH_MAX_DIGESTSIZE];
 850
 851                 if (WARN_ON(size > sizeof(digest))) {
 852                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 853                         goto err;
 854                 }
 855                 r = crypto_shash_final(desc, digest);
 856                 if (unlikely(r < 0)) {
 857                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 858                         goto err;
 859                 }
 860                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 861         }
 862
 863         return;
 864 err:
 865         memset(result, 0, JOURNAL_MAC_SIZE);
 866 }
 867
 868 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 869 {
 870         __u8 result[JOURNAL_MAC_SIZE];
 871         unsigned j;
 872
 873         if (!ic->journal_mac)
 874                 return;
 875
 876         section_mac(ic, section, result);
 877
 878         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 879                 struct journal_sector *js = access_journal(ic, section, j);
 880
 881                 if (likely(wr))
 882                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 883                 else {
 884                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR)) {
 885                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 886                                 dm_audit_log_target(DM_MSG_PREFIX, "mac-journal", ic->ti, 0);
 887                         }
 888                 }
 889         }
 890 }
 891
 892 static void complete_journal_op(void *context)
 893 {
 894         struct journal_completion *comp = context;
 895         BUG_ON(!atomic_read(&comp->in_flight));
 896         if (likely(atomic_dec_and_test(&comp->in_flight)))
 897                 complete(&comp->comp);
 898 }
 899
 900 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 901                         unsigned n_sections, struct journal_completion *comp)
 902 {
 903         struct async_submit_ctl submit;
 904         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 905         unsigned pl_index, pl_offset, section_index;
 906         struct page_list *source_pl, *target_pl;
 907
 908         if (likely(encrypt)) {
 909                 source_pl = ic->journal;
 910                 target_pl = ic->journal_io;
 911         } else {
 912                 source_pl = ic->journal_io;
 913                 target_pl = ic->journal;
 914         }
 915
 916         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 917
 918         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 919
 920         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 921
 922         section_index = pl_index;
 923
 924         do {
 925                 size_t this_step;
 926                 struct page *src_pages[2];
 927                 struct page *dst_page;
 928
 929                 while (unlikely(pl_index == section_index)) {
 930                         unsigned dummy;
 931                         if (likely(encrypt))
 932                                 rw_section_mac(ic, section, true);
 933                         section++;
 934                         n_sections--;
 935                         if (!n_sections)
 936                                 break;
 937                         page_list_location(ic, section, 0, &section_index, &dummy);
 938                 }
 939
 940                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 941                 dst_page = target_pl[pl_index].page;
 942                 src_pages[0] = source_pl[pl_index].page;
 943                 src_pages[1] = ic->journal_xor[pl_index].page;
 944
 945                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 946
 947                 pl_index++;
 948                 pl_offset = 0;
 949                 n_bytes -= this_step;
 950         } while (n_bytes);
 951
 952         BUG_ON(n_sections);
 953
 954         async_tx_issue_pending_all();
 955 }
 956
 957 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 958 {
 959         struct journal_completion *comp = req->data;
 960         if (unlikely(err)) {
 961                 if (likely(err == -EINPROGRESS)) {
 962                         complete(&comp->ic->crypto_backoff);
 963                         return;
 964                 }
 965                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 966         }
 967         complete_journal_op(comp);
 968 }
 969
 970 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 971 {
 972         int r;
 973         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 974                                       complete_journal_encrypt, comp);
 975         if (likely(encrypt))
 976                 r = crypto_skcipher_encrypt(req);
 977         else
 978                 r = crypto_skcipher_decrypt(req);
 979         if (likely(!r))
 980                 return false;
 981         if (likely(r == -EINPROGRESS))
 982                 return true;
 983         if (likely(r == -EBUSY)) {
 984                 wait_for_completion(&comp->ic->crypto_backoff);
 985                 reinit_completion(&comp->ic->crypto_backoff);
 986                 return true;
 987         }
 988         dm_integrity_io_error(comp->ic, "encrypt", r);
 989         return false;
 990 }
 991
 992 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 993                           unsigned n_sections, struct journal_completion *comp)
 994 {
 995         struct scatterlist **source_sg;
 996         struct scatterlist **target_sg;
 997
 998         atomic_add(2, &comp->in_flight);
 999
1000         if (likely(encrypt)) {
1001                 source_sg = ic->journal_scatterlist;
1002                 target_sg = ic->journal_io_scatterlist;
1003         } else {
1004                 source_sg = ic->journal_io_scatterlist;
1005                 target_sg = ic->journal_scatterlist;
1006         }
1007
1008         do {
1009                 struct skcipher_request *req;
1010                 unsigned ivsize;
1011                 char *iv;
1012
1013                 if (likely(encrypt))
1014                         rw_section_mac(ic, section, true);
1015
1016                 req = ic->sk_requests[section];
1017                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1018                 iv = req->iv;
1019
1020                 memcpy(iv, iv + ivsize, ivsize);
1021
1022                 req->src = source_sg[section];
1023                 req->dst = target_sg[section];
1024
1025                 if (unlikely(do_crypt(encrypt, req, comp)))
1026                         atomic_inc(&comp->in_flight);
1027
1028                 section++;
1029                 n_sections--;
1030         } while (n_sections);
1031
1032         atomic_dec(&comp->in_flight);
1033         complete_journal_op(comp);
1034 }
1035
1036 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
1037                             unsigned n_sections, struct journal_completion *comp)
1038 {
1039         if (ic->journal_xor)
1040                 return xor_journal(ic, encrypt, section, n_sections, comp);
1041         else
1042                 return crypt_journal(ic, encrypt, section, n_sections, comp);
1043 }
1044
1045 static void complete_journal_io(unsigned long error, void *context)
1046 {
1047         struct journal_completion *comp = context;
1048         if (unlikely(error != 0))
1049                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1050         complete_journal_op(comp);
1051 }
1052
1053 static void rw_journal_sectors(struct dm_integrity_c *ic, blk_opf_t opf,
1054                                unsigned sector, unsigned n_sectors,
1055                                struct journal_completion *comp)
1056 {
1057         struct dm_io_request io_req;
1058         struct dm_io_region io_loc;
1059         unsigned pl_index, pl_offset;
1060         int r;
1061
1062         if (unlikely(dm_integrity_failed(ic))) {
1063                 if (comp)
1064                         complete_journal_io(-1UL, comp);
1065                 return;
1066         }
1067
1068         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1069         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1070
1071         io_req.bi_opf = opf;
1072         io_req.mem.type = DM_IO_PAGE_LIST;
1073         if (ic->journal_io)
1074                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1075         else
1076                 io_req.mem.ptr.pl = &ic->journal[pl_index];
1077         io_req.mem.offset = pl_offset;
1078         if (likely(comp != NULL)) {
1079                 io_req.notify.fn = complete_journal_io;
1080                 io_req.notify.context = comp;
1081         } else {
1082                 io_req.notify.fn = NULL;
1083         }
1084         io_req.client = ic->io;
1085         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1086         io_loc.sector = ic->start + SB_SECTORS + sector;
1087         io_loc.count = n_sectors;
1088
1089         r = dm_io(&io_req, 1, &io_loc, NULL);
1090         if (unlikely(r)) {
1091                 dm_integrity_io_error(ic, (opf & REQ_OP_MASK) == REQ_OP_READ ?
1092                                       "reading journal" : "writing journal", r);
1093                 if (comp) {
1094                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1095                         complete_journal_io(-1UL, comp);
1096                 }
1097         }
1098 }
1099
1100 static void rw_journal(struct dm_integrity_c *ic, blk_opf_t opf,
1101                        unsigned section, unsigned n_sections,
1102                        struct journal_completion *comp)
1103 {
1104         unsigned sector, n_sectors;
1105
1106         sector = section * ic->journal_section_sectors;
1107         n_sectors = n_sections * ic->journal_section_sectors;
1108
1109         rw_journal_sectors(ic, opf, sector, n_sectors, comp);
1110 }
1111
1112 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1113 {
1114         struct journal_completion io_comp;
1115         struct journal_completion crypt_comp_1;
1116         struct journal_completion crypt_comp_2;
1117         unsigned i;
1118
1119         io_comp.ic = ic;
1120         init_completion(&io_comp.comp);
1121
1122         if (commit_start + commit_sections <= ic->journal_sections) {
1123                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1124                 if (ic->journal_io) {
1125                         crypt_comp_1.ic = ic;
1126                         init_completion(&crypt_comp_1.comp);
1127                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1128                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1129                         wait_for_completion_io(&crypt_comp_1.comp);
1130                 } else {
1131                         for (i = 0; i < commit_sections; i++)
1132                                 rw_section_mac(ic, commit_start + i, true);
1133                 }
1134                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, commit_start,
1135                            commit_sections, &io_comp);
1136         } else {
1137                 unsigned to_end;
1138                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1139                 to_end = ic->journal_sections - commit_start;
1140                 if (ic->journal_io) {
1141                         crypt_comp_1.ic = ic;
1142                         init_completion(&crypt_comp_1.comp);
1143                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1144                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1145                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
1146                                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA,
1147                                            commit_start, to_end, &io_comp);
1148                                 reinit_completion(&crypt_comp_1.comp);
1149                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1150                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1151                                 wait_for_completion_io(&crypt_comp_1.comp);
1152                         } else {
1153                                 crypt_comp_2.ic = ic;
1154                                 init_completion(&crypt_comp_2.comp);
1155                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1156                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1157                                 wait_for_completion_io(&crypt_comp_1.comp);
1158                                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1159                                 wait_for_completion_io(&crypt_comp_2.comp);
1160                         }
1161                 } else {
1162                         for (i = 0; i < to_end; i++)
1163                                 rw_section_mac(ic, commit_start + i, true);
1164                         rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1165                         for (i = 0; i < commit_sections - to_end; i++)
1166                                 rw_section_mac(ic, i, true);
1167                 }
1168                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, 0, commit_sections - to_end, &io_comp);
1169         }
1170
1171         wait_for_completion_io(&io_comp.comp);
1172 }
1173
1174 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1175                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1176 {
1177         struct dm_io_request io_req;
1178         struct dm_io_region io_loc;
1179         int r;
1180         unsigned sector, pl_index, pl_offset;
1181
1182         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1183
1184         if (unlikely(dm_integrity_failed(ic))) {
1185                 fn(-1UL, data);
1186                 return;
1187         }
1188
1189         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1190
1191         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1192         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1193
1194         io_req.bi_opf = REQ_OP_WRITE;
1195         io_req.mem.type = DM_IO_PAGE_LIST;
1196         io_req.mem.ptr.pl = &ic->journal[pl_index];
1197         io_req.mem.offset = pl_offset;
1198         io_req.notify.fn = fn;
1199         io_req.notify.context = data;
1200         io_req.client = ic->io;
1201         io_loc.bdev = ic->dev->bdev;
1202         io_loc.sector = target;
1203         io_loc.count = n_sectors;
1204
1205         r = dm_io(&io_req, 1, &io_loc, NULL);
1206         if (unlikely(r)) {
1207                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1208                 fn(-1UL, data);
1209         }
1210 }
1211
1212 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1213 {
1214         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1215                range1->logical_sector + range1->n_sectors > range2->logical_sector;
1216 }
1217
1218 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1219 {
1220         struct rb_node **n = &ic->in_progress.rb_node;
1221         struct rb_node *parent;
1222
1223         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1224
1225         if (likely(check_waiting)) {
1226                 struct dm_integrity_range *range;
1227                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1228                         if (unlikely(ranges_overlap(range, new_range)))
1229                                 return false;
1230                 }
1231         }
1232
1233         parent = NULL;
1234
1235         while (*n) {
1236                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1237
1238                 parent = *n;
1239                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1240                         n = &range->node.rb_left;
1241                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1242                         n = &range->node.rb_right;
1243                 } else {
1244                         return false;
1245                 }
1246         }
1247
1248         rb_link_node(&new_range->node, parent, n);
1249         rb_insert_color(&new_range->node, &ic->in_progress);
1250
1251         return true;
1252 }
1253
1254 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1255 {
1256         rb_erase(&range->node, &ic->in_progress);
1257         while (unlikely(!list_empty(&ic->wait_list))) {
1258                 struct dm_integrity_range *last_range =
1259                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1260                 struct task_struct *last_range_task;
1261                 last_range_task = last_range->task;
1262                 list_del(&last_range->wait_entry);
1263                 if (!add_new_range(ic, last_range, false)) {
1264                         last_range->task = last_range_task;
1265                         list_add(&last_range->wait_entry, &ic->wait_list);
1266                         break;
1267                 }
1268                 last_range->waiting = false;
1269                 wake_up_process(last_range_task);
1270         }
1271 }
1272
1273 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1274 {
1275         unsigned long flags;
1276
1277         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1278         remove_range_unlocked(ic, range);
1279         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1280 }
1281
1282 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1283 {
1284         new_range->waiting = true;
1285         list_add_tail(&new_range->wait_entry, &ic->wait_list);
1286         new_range->task = current;
1287         do {
1288                 __set_current_state(TASK_UNINTERRUPTIBLE);
1289                 spin_unlock_irq(&ic->endio_wait.lock);
1290                 io_schedule();
1291                 spin_lock_irq(&ic->endio_wait.lock);
1292         } while (unlikely(new_range->waiting));
1293 }
1294
1295 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1296 {
1297         if (unlikely(!add_new_range(ic, new_range, true)))
1298                 wait_and_add_new_range(ic, new_range);
1299 }
1300
1301 static void init_journal_node(struct journal_node *node)
1302 {
1303         RB_CLEAR_NODE(&node->node);
1304         node->sector = (sector_t)-1;
1305 }
1306
1307 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1308 {
1309         struct rb_node **link;
1310         struct rb_node *parent;
1311
1312         node->sector = sector;
1313         BUG_ON(!RB_EMPTY_NODE(&node->node));
1314
1315         link = &ic->journal_tree_root.rb_node;
1316         parent = NULL;
1317
1318         while (*link) {
1319                 struct journal_node *j;
1320                 parent = *link;
1321                 j = container_of(parent, struct journal_node, node);
1322                 if (sector < j->sector)
1323                         link = &j->node.rb_left;
1324                 else
1325                         link = &j->node.rb_right;
1326         }
1327
1328         rb_link_node(&node->node, parent, link);
1329         rb_insert_color(&node->node, &ic->journal_tree_root);
1330 }
1331
1332 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1333 {
1334         BUG_ON(RB_EMPTY_NODE(&node->node));
1335         rb_erase(&node->node, &ic->journal_tree_root);
1336         init_journal_node(node);
1337 }
1338
1339 #define NOT_FOUND       (-1U)
1340
1341 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1342 {
1343         struct rb_node *n = ic->journal_tree_root.rb_node;
1344         unsigned found = NOT_FOUND;
1345         *next_sector = (sector_t)-1;
1346         while (n) {
1347                 struct journal_node *j = container_of(n, struct journal_node, node);
1348                 if (sector == j->sector) {
1349                         found = j - ic->journal_tree;
1350                 }
1351                 if (sector < j->sector) {
1352                         *next_sector = j->sector;
1353                         n = j->node.rb_left;
1354                 } else {
1355                         n = j->node.rb_right;
1356                 }
1357         }
1358
1359         return found;
1360 }
1361
1362 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1363 {
1364         struct journal_node *node, *next_node;
1365         struct rb_node *next;
1366
1367         if (unlikely(pos >= ic->journal_entries))
1368                 return false;
1369         node = &ic->journal_tree[pos];
1370         if (unlikely(RB_EMPTY_NODE(&node->node)))
1371                 return false;
1372         if (unlikely(node->sector != sector))
1373                 return false;
1374
1375         next = rb_next(&node->node);
1376         if (unlikely(!next))
1377                 return true;
1378
1379         next_node = container_of(next, struct journal_node, node);
1380         return next_node->sector != sector;
1381 }
1382
1383 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1384 {
1385         struct rb_node *next;
1386         struct journal_node *next_node;
1387         unsigned next_section;
1388
1389         BUG_ON(RB_EMPTY_NODE(&node->node));
1390
1391         next = rb_next(&node->node);
1392         if (unlikely(!next))
1393                 return false;
1394
1395         next_node = container_of(next, struct journal_node, node);
1396
1397         if (next_node->sector != node->sector)
1398                 return false;
1399
1400         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1401         if (next_section >= ic->committed_section &&
1402             next_section < ic->committed_section + ic->n_committed_sections)
1403                 return true;
1404         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1405                 return true;
1406
1407         return false;
1408 }
1409
1410 #define TAG_READ        0
1411 #define TAG_WRITE       1
1412 #define TAG_CMP         2
1413
1414 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1415                                unsigned *metadata_offset, unsigned total_size, int op)
1416 {
1417 #define MAY_BE_FILLER           1
1418 #define MAY_BE_HASH             2
1419         unsigned hash_offset = 0;
1420         unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1421
1422         do {
1423                 unsigned char *data, *dp;
1424                 struct dm_buffer *b;
1425                 unsigned to_copy;
1426                 int r;
1427
1428                 r = dm_integrity_failed(ic);
1429                 if (unlikely(r))
1430                         return r;
1431
1432                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1433                 if (IS_ERR(data))
1434                         return PTR_ERR(data);
1435
1436                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1437                 dp = data + *metadata_offset;
1438                 if (op == TAG_READ) {
1439                         memcpy(tag, dp, to_copy);
1440                 } else if (op == TAG_WRITE) {
1441                         if (memcmp(dp, tag, to_copy)) {
1442                                 memcpy(dp, tag, to_copy);
1443                                 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1444                         }
1445                 } else {
1446                         /* e.g.: op == TAG_CMP */
1447
1448                         if (likely(is_power_of_2(ic->tag_size))) {
1449                                 if (unlikely(memcmp(dp, tag, to_copy)))
1450                                         if (unlikely(!ic->discard) ||
1451                                             unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1452                                                 goto thorough_test;
1453                                 }
1454                         } else {
1455                                 unsigned i, ts;
1456 thorough_test:
1457                                 ts = total_size;
1458
1459                                 for (i = 0; i < to_copy; i++, ts--) {
1460                                         if (unlikely(dp[i] != tag[i]))
1461                                                 may_be &= ~MAY_BE_HASH;
1462                                         if (likely(dp[i] != DISCARD_FILLER))
1463                                                 may_be &= ~MAY_BE_FILLER;
1464                                         hash_offset++;
1465                                         if (unlikely(hash_offset == ic->tag_size)) {
1466                                                 if (unlikely(!may_be)) {
1467                                                         dm_bufio_release(b);
1468                                                         return ts;
1469                                                 }
1470                                                 hash_offset = 0;
1471                                                 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1472                                         }
1473                                 }
1474                         }
1475                 }
1476                 dm_bufio_release(b);
1477
1478                 tag += to_copy;
1479                 *metadata_offset += to_copy;
1480                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1481                         (*metadata_block)++;
1482                         *metadata_offset = 0;
1483                 }
1484
1485                 if (unlikely(!is_power_of_2(ic->tag_size))) {
1486                         hash_offset = (hash_offset + to_copy) % ic->tag_size;
1487                 }
1488
1489                 total_size -= to_copy;
1490         } while (unlikely(total_size));
1491
1492         return 0;
1493 #undef MAY_BE_FILLER
1494 #undef MAY_BE_HASH
1495 }
1496
1497 struct flush_request {
1498         struct dm_io_request io_req;
1499         struct dm_io_region io_reg;
1500         struct dm_integrity_c *ic;
1501         struct completion comp;
1502 };
1503
1504 static void flush_notify(unsigned long error, void *fr_)
1505 {
1506         struct flush_request *fr = fr_;
1507         if (unlikely(error != 0))
1508                 dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1509         complete(&fr->comp);
1510 }
1511
1512 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1513 {
1514         int r;
1515
1516         struct flush_request fr;
1517
1518         if (!ic->meta_dev)
1519                 flush_data = false;
1520         if (flush_data) {
1521                 fr.io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC,
1522                 fr.io_req.mem.type = DM_IO_KMEM,
1523                 fr.io_req.mem.ptr.addr = NULL,
1524                 fr.io_req.notify.fn = flush_notify,
1525                 fr.io_req.notify.context = &fr;
1526                 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1527                 fr.io_reg.bdev = ic->dev->bdev,
1528                 fr.io_reg.sector = 0,
1529                 fr.io_reg.count = 0,
1530                 fr.ic = ic;
1531                 init_completion(&fr.comp);
1532                 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1533                 BUG_ON(r);
1534         }
1535
1536         r = dm_bufio_write_dirty_buffers(ic->bufio);
1537         if (unlikely(r))
1538                 dm_integrity_io_error(ic, "writing tags", r);
1539
1540         if (flush_data)
1541                 wait_for_completion(&fr.comp);
1542 }
1543
1544 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1545 {
1546         DECLARE_WAITQUEUE(wait, current);
1547         __add_wait_queue(&ic->endio_wait, &wait);
1548         __set_current_state(TASK_UNINTERRUPTIBLE);
1549         spin_unlock_irq(&ic->endio_wait.lock);
1550         io_schedule();
1551         spin_lock_irq(&ic->endio_wait.lock);
1552         __remove_wait_queue(&ic->endio_wait, &wait);
1553 }
1554
1555 static void autocommit_fn(struct timer_list *t)
1556 {
1557         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1558
1559         if (likely(!dm_integrity_failed(ic)))
1560                 queue_work(ic->commit_wq, &ic->commit_work);
1561 }
1562
1563 static void schedule_autocommit(struct dm_integrity_c *ic)
1564 {
1565         if (!timer_pending(&ic->autocommit_timer))
1566                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1567 }
1568
1569 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1570 {
1571         struct bio *bio;
1572         unsigned long flags;
1573
1574         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1575         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1576         bio_list_add(&ic->flush_bio_list, bio);
1577         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1578
1579         queue_work(ic->commit_wq, &ic->commit_work);
1580 }
1581
1582 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1583 {
1584         int r = dm_integrity_failed(ic);
1585         if (unlikely(r) && !bio->bi_status)
1586                 bio->bi_status = errno_to_blk_status(r);
1587         if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1588                 unsigned long flags;
1589                 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1590                 bio_list_add(&ic->synchronous_bios, bio);
1591                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1592                 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1593                 return;
1594         }
1595         bio_endio(bio);
1596 }
1597
1598 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1599 {
1600         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1601
1602         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1603                 submit_flush_bio(ic, dio);
1604         else
1605                 do_endio(ic, bio);
1606 }
1607
1608 static void dec_in_flight(struct dm_integrity_io *dio)
1609 {
1610         if (atomic_dec_and_test(&dio->in_flight)) {
1611                 struct dm_integrity_c *ic = dio->ic;
1612                 struct bio *bio;
1613
1614                 remove_range(ic, &dio->range);
1615
1616                 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1617                         schedule_autocommit(ic);
1618
1619                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1620
1621                 if (unlikely(dio->bi_status) && !bio->bi_status)
1622                         bio->bi_status = dio->bi_status;
1623                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1624                         dio->range.logical_sector += dio->range.n_sectors;
1625                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1626                         INIT_WORK(&dio->work, integrity_bio_wait);
1627                         queue_work(ic->offload_wq, &dio->work);
1628                         return;
1629                 }
1630                 do_endio_flush(ic, dio);
1631         }
1632 }
1633
1634 static void integrity_end_io(struct bio *bio)
1635 {
1636         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1637
1638         dm_bio_restore(&dio->bio_details, bio);
1639         if (bio->bi_integrity)
1640                 bio->bi_opf |= REQ_INTEGRITY;
1641
1642         if (dio->completion)
1643                 complete(dio->completion);
1644
1645         dec_in_flight(dio);
1646 }
1647
1648 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1649                                       const char *data, char *result)
1650 {
1651         __le64 sector_le = cpu_to_le64(sector);
1652         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1653         int r;
1654         unsigned digest_size;
1655
1656         req->tfm = ic->internal_hash;
1657
1658         r = crypto_shash_init(req);
1659         if (unlikely(r < 0)) {
1660                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1661                 goto failed;
1662         }
1663
1664         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1665                 r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1666                 if (unlikely(r < 0)) {
1667                         dm_integrity_io_error(ic, "crypto_shash_update", r);
1668                         goto failed;
1669                 }
1670         }
1671
1672         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1673         if (unlikely(r < 0)) {
1674                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1675                 goto failed;
1676         }
1677
1678         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1679         if (unlikely(r < 0)) {
1680                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1681                 goto failed;
1682         }
1683
1684         r = crypto_shash_final(req, result);
1685         if (unlikely(r < 0)) {
1686                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1687                 goto failed;
1688         }
1689
1690         digest_size = crypto_shash_digestsize(ic->internal_hash);
1691         if (unlikely(digest_size < ic->tag_size))
1692                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1693
1694         return;
1695
1696 failed:
1697         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1698         get_random_bytes(result, ic->tag_size);
1699 }
1700
1701 static void integrity_metadata(struct work_struct *w)
1702 {
1703         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1704         struct dm_integrity_c *ic = dio->ic;
1705
1706         int r;
1707
1708         if (ic->internal_hash) {
1709                 struct bvec_iter iter;
1710                 struct bio_vec bv;
1711                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1712                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1713                 char *checksums;
1714                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1715                 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1716                 sector_t sector;
1717                 unsigned sectors_to_process;
1718
1719                 if (unlikely(ic->mode == 'R'))
1720                         goto skip_io;
1721
1722                 if (likely(dio->op != REQ_OP_DISCARD))
1723                         checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1724                                             GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1725                 else
1726                         checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1727                 if (!checksums) {
1728                         checksums = checksums_onstack;
1729                         if (WARN_ON(extra_space &&
1730                                     digest_size > sizeof(checksums_onstack))) {
1731                                 r = -EINVAL;
1732                                 goto error;
1733                         }
1734                 }
1735
1736                 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1737                         sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1738                         unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1739                         unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1740                         unsigned max_blocks = max_size / ic->tag_size;
1741                         memset(checksums, DISCARD_FILLER, max_size);
1742
1743                         while (bi_size) {
1744                                 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1745                                 this_step_blocks = min(this_step_blocks, max_blocks);
1746                                 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1747                                                         this_step_blocks * ic->tag_size, TAG_WRITE);
1748                                 if (unlikely(r)) {
1749                                         if (likely(checksums != checksums_onstack))
1750                                                 kfree(checksums);
1751                                         goto error;
1752                                 }
1753
1754                                 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1755                                         printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1756                                         printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1757                                         BUG();
1758                                 }*/
1759                                 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1760                                 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1761                         }
1762
1763                         if (likely(checksums != checksums_onstack))
1764                                 kfree(checksums);
1765                         goto skip_io;
1766                 }
1767
1768                 sector = dio->range.logical_sector;
1769                 sectors_to_process = dio->range.n_sectors;
1770
1771                 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1772                         unsigned pos;
1773                         char *mem, *checksums_ptr;
1774
1775 again:
1776                         mem = bvec_kmap_local(&bv);
1777                         pos = 0;
1778                         checksums_ptr = checksums;
1779                         do {
1780                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1781                                 checksums_ptr += ic->tag_size;
1782                                 sectors_to_process -= ic->sectors_per_block;
1783                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1784                                 sector += ic->sectors_per_block;
1785                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1786                         kunmap_local(mem);
1787
1788                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1789                                                 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1790                         if (unlikely(r)) {
1791                                 if (r > 0) {
1792                                         sector_t s;
1793
1794                                         s = sector - ((r + ic->tag_size - 1) / ic->tag_size);
1795                                         DMERR_LIMIT("%pg: Checksum failed at sector 0x%llx",
1796                                                     bio->bi_bdev, s);
1797                                         r = -EILSEQ;
1798                                         atomic64_inc(&ic->number_of_mismatches);
1799                                         dm_audit_log_bio(DM_MSG_PREFIX, "integrity-checksum",
1800                                                          bio, s, 0);
1801                                 }
1802                                 if (likely(checksums != checksums_onstack))
1803                                         kfree(checksums);
1804                                 goto error;
1805                         }
1806
1807                         if (!sectors_to_process)
1808                                 break;
1809
1810                         if (unlikely(pos < bv.bv_len)) {
1811                                 bv.bv_offset += pos;
1812                                 bv.bv_len -= pos;
1813                                 goto again;
1814                         }
1815                 }
1816
1817                 if (likely(checksums != checksums_onstack))
1818                         kfree(checksums);
1819         } else {
1820                 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1821
1822                 if (bip) {
1823                         struct bio_vec biv;
1824                         struct bvec_iter iter;
1825                         unsigned data_to_process = dio->range.n_sectors;
1826                         sector_to_block(ic, data_to_process);
1827                         data_to_process *= ic->tag_size;
1828
1829                         bip_for_each_vec(biv, bip, iter) {
1830                                 unsigned char *tag;
1831                                 unsigned this_len;
1832
1833                                 BUG_ON(PageHighMem(biv.bv_page));
1834                                 tag = bvec_virt(&biv);
1835                                 this_len = min(biv.bv_len, data_to_process);
1836                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1837                                                         this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1838                                 if (unlikely(r))
1839                                         goto error;
1840                                 data_to_process -= this_len;
1841                                 if (!data_to_process)
1842                                         break;
1843                         }
1844                 }
1845         }
1846 skip_io:
1847         dec_in_flight(dio);
1848         return;
1849 error:
1850         dio->bi_status = errno_to_blk_status(r);
1851         dec_in_flight(dio);
1852 }
1853
1854 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1855 {
1856         struct dm_integrity_c *ic = ti->private;
1857         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1858         struct bio_integrity_payload *bip;
1859
1860         sector_t area, offset;
1861
1862         dio->ic = ic;
1863         dio->bi_status = 0;
1864         dio->op = bio_op(bio);
1865
1866         if (unlikely(dio->op == REQ_OP_DISCARD)) {
1867                 if (ti->max_io_len) {
1868                         sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1869                         unsigned log2_max_io_len = __fls(ti->max_io_len);
1870                         sector_t start_boundary = sec >> log2_max_io_len;
1871                         sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1872                         if (start_boundary < end_boundary) {
1873                                 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1874                                 dm_accept_partial_bio(bio, len);
1875                         }
1876                 }
1877         }
1878
1879         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1880                 submit_flush_bio(ic, dio);
1881                 return DM_MAPIO_SUBMITTED;
1882         }
1883
1884         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1885         dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1886         if (unlikely(dio->fua)) {
1887                 /*
1888                  * Don't pass down the FUA flag because we have to flush
1889                  * disk cache anyway.
1890                  */
1891                 bio->bi_opf &= ~REQ_FUA;
1892         }
1893         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1894                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1895                       dio->range.logical_sector, bio_sectors(bio),
1896                       ic->provided_data_sectors);
1897                 return DM_MAPIO_KILL;
1898         }
1899         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1900                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1901                       ic->sectors_per_block,
1902                       dio->range.logical_sector, bio_sectors(bio));
1903                 return DM_MAPIO_KILL;
1904         }
1905
1906         if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1907                 struct bvec_iter iter;
1908                 struct bio_vec bv;
1909                 bio_for_each_segment(bv, bio, iter) {
1910                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1911                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1912                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1913                                 return DM_MAPIO_KILL;
1914                         }
1915                 }
1916         }
1917
1918         bip = bio_integrity(bio);
1919         if (!ic->internal_hash) {
1920                 if (bip) {
1921                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1922                         if (ic->log2_tag_size >= 0)
1923                                 wanted_tag_size <<= ic->log2_tag_size;
1924                         else
1925                                 wanted_tag_size *= ic->tag_size;
1926                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1927                                 DMERR("Invalid integrity data size %u, expected %u",
1928                                       bip->bip_iter.bi_size, wanted_tag_size);
1929                                 return DM_MAPIO_KILL;
1930                         }
1931                 }
1932         } else {
1933                 if (unlikely(bip != NULL)) {
1934                         DMERR("Unexpected integrity data when using internal hash");
1935                         return DM_MAPIO_KILL;
1936                 }
1937         }
1938
1939         if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1940                 return DM_MAPIO_KILL;
1941
1942         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1943         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1944         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1945
1946         dm_integrity_map_continue(dio, true);
1947         return DM_MAPIO_SUBMITTED;
1948 }
1949
1950 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1951                                  unsigned journal_section, unsigned journal_entry)
1952 {
1953         struct dm_integrity_c *ic = dio->ic;
1954         sector_t logical_sector;
1955         unsigned n_sectors;
1956
1957         logical_sector = dio->range.logical_sector;
1958         n_sectors = dio->range.n_sectors;
1959         do {
1960                 struct bio_vec bv = bio_iovec(bio);
1961                 char *mem;
1962
1963                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1964                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1965                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1966                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1967 retry_kmap:
1968                 mem = kmap_local_page(bv.bv_page);
1969                 if (likely(dio->op == REQ_OP_WRITE))
1970                         flush_dcache_page(bv.bv_page);
1971
1972                 do {
1973                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1974
1975                         if (unlikely(dio->op == REQ_OP_READ)) {
1976                                 struct journal_sector *js;
1977                                 char *mem_ptr;
1978                                 unsigned s;
1979
1980                                 if (unlikely(journal_entry_is_inprogress(je))) {
1981                                         flush_dcache_page(bv.bv_page);
1982                                         kunmap_local(mem);
1983
1984                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1985                                         goto retry_kmap;
1986                                 }
1987                                 smp_rmb();
1988                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1989                                 js = access_journal_data(ic, journal_section, journal_entry);
1990                                 mem_ptr = mem + bv.bv_offset;
1991                                 s = 0;
1992                                 do {
1993                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1994                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1995                                         js++;
1996                                         mem_ptr += 1 << SECTOR_SHIFT;
1997                                 } while (++s < ic->sectors_per_block);
1998 #ifdef INTERNAL_VERIFY
1999                                 if (ic->internal_hash) {
2000                                         char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2001
2002                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
2003                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
2004                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
2005                                                             logical_sector);
2006                                                 dm_audit_log_bio(DM_MSG_PREFIX, "journal-checksum",
2007                                                                  bio, logical_sector, 0);
2008                                         }
2009                                 }
2010 #endif
2011                         }
2012
2013                         if (!ic->internal_hash) {
2014                                 struct bio_integrity_payload *bip = bio_integrity(bio);
2015                                 unsigned tag_todo = ic->tag_size;
2016                                 char *tag_ptr = journal_entry_tag(ic, je);
2017
2018                                 if (bip) do {
2019                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2020                                         unsigned tag_now = min(biv.bv_len, tag_todo);
2021                                         char *tag_addr;
2022                                         BUG_ON(PageHighMem(biv.bv_page));
2023                                         tag_addr = bvec_virt(&biv);
2024                                         if (likely(dio->op == REQ_OP_WRITE))
2025                                                 memcpy(tag_ptr, tag_addr, tag_now);
2026                                         else
2027                                                 memcpy(tag_addr, tag_ptr, tag_now);
2028                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2029                                         tag_ptr += tag_now;
2030                                         tag_todo -= tag_now;
2031                                 } while (unlikely(tag_todo)); else {
2032                                         if (likely(dio->op == REQ_OP_WRITE))
2033                                                 memset(tag_ptr, 0, tag_todo);
2034                                 }
2035                         }
2036
2037                         if (likely(dio->op == REQ_OP_WRITE)) {
2038                                 struct journal_sector *js;
2039                                 unsigned s;
2040
2041                                 js = access_journal_data(ic, journal_section, journal_entry);
2042                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2043
2044                                 s = 0;
2045                                 do {
2046                                         je->last_bytes[s] = js[s].commit_id;
2047                                 } while (++s < ic->sectors_per_block);
2048
2049                                 if (ic->internal_hash) {
2050                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
2051                                         if (unlikely(digest_size > ic->tag_size)) {
2052                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
2053                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2054                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2055                                         } else
2056                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2057                                 }
2058
2059                                 journal_entry_set_sector(je, logical_sector);
2060                         }
2061                         logical_sector += ic->sectors_per_block;
2062
2063                         journal_entry++;
2064                         if (unlikely(journal_entry == ic->journal_section_entries)) {
2065                                 journal_entry = 0;
2066                                 journal_section++;
2067                                 wraparound_section(ic, &journal_section);
2068                         }
2069
2070                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2071                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2072
2073                 if (unlikely(dio->op == REQ_OP_READ))
2074                         flush_dcache_page(bv.bv_page);
2075                 kunmap_local(mem);
2076         } while (n_sectors);
2077
2078         if (likely(dio->op == REQ_OP_WRITE)) {
2079                 smp_mb();
2080                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2081                         wake_up(&ic->copy_to_journal_wait);
2082                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
2083                         queue_work(ic->commit_wq, &ic->commit_work);
2084                 } else {
2085                         schedule_autocommit(ic);
2086                 }
2087         } else {
2088                 remove_range(ic, &dio->range);
2089         }
2090
2091         if (unlikely(bio->bi_iter.bi_size)) {
2092                 sector_t area, offset;
2093
2094                 dio->range.logical_sector = logical_sector;
2095                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2096                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2097                 return true;
2098         }
2099
2100         return false;
2101 }
2102
2103 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2104 {
2105         struct dm_integrity_c *ic = dio->ic;
2106         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2107         unsigned journal_section, journal_entry;
2108         unsigned journal_read_pos;
2109         struct completion read_comp;
2110         bool discard_retried = false;
2111         bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2112         if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2113                 need_sync_io = true;
2114
2115         if (need_sync_io && from_map) {
2116                 INIT_WORK(&dio->work, integrity_bio_wait);
2117                 queue_work(ic->offload_wq, &dio->work);
2118                 return;
2119         }
2120
2121 lock_retry:
2122         spin_lock_irq(&ic->endio_wait.lock);
2123 retry:
2124         if (unlikely(dm_integrity_failed(ic))) {
2125                 spin_unlock_irq(&ic->endio_wait.lock);
2126                 do_endio(ic, bio);
2127                 return;
2128         }
2129         dio->range.n_sectors = bio_sectors(bio);
2130         journal_read_pos = NOT_FOUND;
2131         if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2132                 if (dio->op == REQ_OP_WRITE) {
2133                         unsigned next_entry, i, pos;
2134                         unsigned ws, we, range_sectors;
2135
2136                         dio->range.n_sectors = min(dio->range.n_sectors,
2137                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2138                         if (unlikely(!dio->range.n_sectors)) {
2139                                 if (from_map)
2140                                         goto offload_to_thread;
2141                                 sleep_on_endio_wait(ic);
2142                                 goto retry;
2143                         }
2144                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2145                         ic->free_sectors -= range_sectors;
2146                         journal_section = ic->free_section;
2147                         journal_entry = ic->free_section_entry;
2148
2149                         next_entry = ic->free_section_entry + range_sectors;
2150                         ic->free_section_entry = next_entry % ic->journal_section_entries;
2151                         ic->free_section += next_entry / ic->journal_section_entries;
2152                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2153                         wraparound_section(ic, &ic->free_section);
2154
2155                         pos = journal_section * ic->journal_section_entries + journal_entry;
2156                         ws = journal_section;
2157                         we = journal_entry;
2158                         i = 0;
2159                         do {
2160                                 struct journal_entry *je;
2161
2162                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2163                                 pos++;
2164                                 if (unlikely(pos >= ic->journal_entries))
2165                                         pos = 0;
2166
2167                                 je = access_journal_entry(ic, ws, we);
2168                                 BUG_ON(!journal_entry_is_unused(je));
2169                                 journal_entry_set_inprogress(je);
2170                                 we++;
2171                                 if (unlikely(we == ic->journal_section_entries)) {
2172                                         we = 0;
2173                                         ws++;
2174                                         wraparound_section(ic, &ws);
2175                                 }
2176                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2177
2178                         spin_unlock_irq(&ic->endio_wait.lock);
2179                         goto journal_read_write;
2180                 } else {
2181                         sector_t next_sector;
2182                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2183                         if (likely(journal_read_pos == NOT_FOUND)) {
2184                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2185                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
2186                         } else {
2187                                 unsigned i;
2188                                 unsigned jp = journal_read_pos + 1;
2189                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2190                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2191                                                 break;
2192                                 }
2193                                 dio->range.n_sectors = i;
2194                         }
2195                 }
2196         }
2197         if (unlikely(!add_new_range(ic, &dio->range, true))) {
2198                 /*
2199                  * We must not sleep in the request routine because it could
2200                  * stall bios on current->bio_list.
2201                  * So, we offload the bio to a workqueue if we have to sleep.
2202                  */
2203                 if (from_map) {
2204 offload_to_thread:
2205                         spin_unlock_irq(&ic->endio_wait.lock);
2206                         INIT_WORK(&dio->work, integrity_bio_wait);
2207                         queue_work(ic->wait_wq, &dio->work);
2208                         return;
2209                 }
2210                 if (journal_read_pos != NOT_FOUND)
2211                         dio->range.n_sectors = ic->sectors_per_block;
2212                 wait_and_add_new_range(ic, &dio->range);
2213                 /*
2214                  * wait_and_add_new_range drops the spinlock, so the journal
2215                  * may have been changed arbitrarily. We need to recheck.
2216                  * To simplify the code, we restrict I/O size to just one block.
2217                  */
2218                 if (journal_read_pos != NOT_FOUND) {
2219                         sector_t next_sector;
2220                         unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2221                         if (unlikely(new_pos != journal_read_pos)) {
2222                                 remove_range_unlocked(ic, &dio->range);
2223                                 goto retry;
2224                         }
2225                 }
2226         }
2227         if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2228                 sector_t next_sector;
2229                 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2230                 if (unlikely(new_pos != NOT_FOUND) ||
2231                     unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2232                         remove_range_unlocked(ic, &dio->range);
2233                         spin_unlock_irq(&ic->endio_wait.lock);
2234                         queue_work(ic->commit_wq, &ic->commit_work);
2235                         flush_workqueue(ic->commit_wq);
2236                         queue_work(ic->writer_wq, &ic->writer_work);
2237                         flush_workqueue(ic->writer_wq);
2238                         discard_retried = true;
2239                         goto lock_retry;
2240                 }
2241         }
2242         spin_unlock_irq(&ic->endio_wait.lock);
2243
2244         if (unlikely(journal_read_pos != NOT_FOUND)) {
2245                 journal_section = journal_read_pos / ic->journal_section_entries;
2246                 journal_entry = journal_read_pos % ic->journal_section_entries;
2247                 goto journal_read_write;
2248         }
2249
2250         if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2251                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2252                                      dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2253                         struct bitmap_block_status *bbs;
2254
2255                         bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2256                         spin_lock(&bbs->bio_queue_lock);
2257                         bio_list_add(&bbs->bio_queue, bio);
2258                         spin_unlock(&bbs->bio_queue_lock);
2259                         queue_work(ic->writer_wq, &bbs->work);
2260                         return;
2261                 }
2262         }
2263
2264         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2265
2266         if (need_sync_io) {
2267                 init_completion(&read_comp);
2268                 dio->completion = &read_comp;
2269         } else
2270                 dio->completion = NULL;
2271
2272         dm_bio_record(&dio->bio_details, bio);
2273         bio_set_dev(bio, ic->dev->bdev);
2274         bio->bi_integrity = NULL;
2275         bio->bi_opf &= ~REQ_INTEGRITY;
2276         bio->bi_end_io = integrity_end_io;
2277         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2278
2279         if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2280                 integrity_metadata(&dio->work);
2281                 dm_integrity_flush_buffers(ic, false);
2282
2283                 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2284                 dio->completion = NULL;
2285
2286                 submit_bio_noacct(bio);
2287
2288                 return;
2289         }
2290
2291         submit_bio_noacct(bio);
2292
2293         if (need_sync_io) {
2294                 wait_for_completion_io(&read_comp);
2295                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2296                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2297                         goto skip_check;
2298                 if (ic->mode == 'B') {
2299                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2300                                              dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2301                                 goto skip_check;
2302                 }
2303
2304                 if (likely(!bio->bi_status))
2305                         integrity_metadata(&dio->work);
2306                 else
2307 skip_check:
2308                         dec_in_flight(dio);
2309
2310         } else {
2311                 INIT_WORK(&dio->work, integrity_metadata);
2312                 queue_work(ic->metadata_wq, &dio->work);
2313         }
2314
2315         return;
2316
2317 journal_read_write:
2318         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2319                 goto lock_retry;
2320
2321         do_endio_flush(ic, dio);
2322 }
2323
2324
2325 static void integrity_bio_wait(struct work_struct *w)
2326 {
2327         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2328
2329         dm_integrity_map_continue(dio, false);
2330 }
2331
2332 static void pad_uncommitted(struct dm_integrity_c *ic)
2333 {
2334         if (ic->free_section_entry) {
2335                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2336                 ic->free_section_entry = 0;
2337                 ic->free_section++;
2338                 wraparound_section(ic, &ic->free_section);
2339                 ic->n_uncommitted_sections++;
2340         }
2341         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2342                     (ic->n_uncommitted_sections + ic->n_committed_sections) *
2343                     ic->journal_section_entries + ic->free_sectors)) {
2344                 DMCRIT("journal_sections %u, journal_section_entries %u, "
2345                        "n_uncommitted_sections %u, n_committed_sections %u, "
2346                        "journal_section_entries %u, free_sectors %u",
2347                        ic->journal_sections, ic->journal_section_entries,
2348                        ic->n_uncommitted_sections, ic->n_committed_sections,
2349                        ic->journal_section_entries, ic->free_sectors);
2350         }
2351 }
2352
2353 static void integrity_commit(struct work_struct *w)
2354 {
2355         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2356         unsigned commit_start, commit_sections;
2357         unsigned i, j, n;
2358         struct bio *flushes;
2359
2360         del_timer(&ic->autocommit_timer);
2361
2362         spin_lock_irq(&ic->endio_wait.lock);
2363         flushes = bio_list_get(&ic->flush_bio_list);
2364         if (unlikely(ic->mode != 'J')) {
2365                 spin_unlock_irq(&ic->endio_wait.lock);
2366                 dm_integrity_flush_buffers(ic, true);
2367                 goto release_flush_bios;
2368         }
2369
2370         pad_uncommitted(ic);
2371         commit_start = ic->uncommitted_section;
2372         commit_sections = ic->n_uncommitted_sections;
2373         spin_unlock_irq(&ic->endio_wait.lock);
2374
2375         if (!commit_sections)
2376                 goto release_flush_bios;
2377
2378         i = commit_start;
2379         for (n = 0; n < commit_sections; n++) {
2380                 for (j = 0; j < ic->journal_section_entries; j++) {
2381                         struct journal_entry *je;
2382                         je = access_journal_entry(ic, i, j);
2383                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2384                 }
2385                 for (j = 0; j < ic->journal_section_sectors; j++) {
2386                         struct journal_sector *js;
2387                         js = access_journal(ic, i, j);
2388                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2389                 }
2390                 i++;
2391                 if (unlikely(i >= ic->journal_sections))
2392                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2393                 wraparound_section(ic, &i);
2394         }
2395         smp_rmb();
2396
2397         write_journal(ic, commit_start, commit_sections);
2398
2399         spin_lock_irq(&ic->endio_wait.lock);
2400         ic->uncommitted_section += commit_sections;
2401         wraparound_section(ic, &ic->uncommitted_section);
2402         ic->n_uncommitted_sections -= commit_sections;
2403         ic->n_committed_sections += commit_sections;
2404         spin_unlock_irq(&ic->endio_wait.lock);
2405
2406         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2407                 queue_work(ic->writer_wq, &ic->writer_work);
2408
2409 release_flush_bios:
2410         while (flushes) {
2411                 struct bio *next = flushes->bi_next;
2412                 flushes->bi_next = NULL;
2413                 do_endio(ic, flushes);
2414                 flushes = next;
2415         }
2416 }
2417
2418 static void complete_copy_from_journal(unsigned long error, void *context)
2419 {
2420         struct journal_io *io = context;
2421         struct journal_completion *comp = io->comp;
2422         struct dm_integrity_c *ic = comp->ic;
2423         remove_range(ic, &io->range);
2424         mempool_free(io, &ic->journal_io_mempool);
2425         if (unlikely(error != 0))
2426                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2427         complete_journal_op(comp);
2428 }
2429
2430 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2431                                struct journal_entry *je)
2432 {
2433         unsigned s = 0;
2434         do {
2435                 js->commit_id = je->last_bytes[s];
2436                 js++;
2437         } while (++s < ic->sectors_per_block);
2438 }
2439
2440 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2441                              unsigned write_sections, bool from_replay)
2442 {
2443         unsigned i, j, n;
2444         struct journal_completion comp;
2445         struct blk_plug plug;
2446
2447         blk_start_plug(&plug);
2448
2449         comp.ic = ic;
2450         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2451         init_completion(&comp.comp);
2452
2453         i = write_start;
2454         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2455 #ifndef INTERNAL_VERIFY
2456                 if (unlikely(from_replay))
2457 #endif
2458                         rw_section_mac(ic, i, false);
2459                 for (j = 0; j < ic->journal_section_entries; j++) {
2460                         struct journal_entry *je = access_journal_entry(ic, i, j);
2461                         sector_t sec, area, offset;
2462                         unsigned k, l, next_loop;
2463                         sector_t metadata_block;
2464                         unsigned metadata_offset;
2465                         struct journal_io *io;
2466
2467                         if (journal_entry_is_unused(je))
2468                                 continue;
2469                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2470                         sec = journal_entry_get_sector(je);
2471                         if (unlikely(from_replay)) {
2472                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2473                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2474                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2475                                 }
2476                                 if (unlikely(sec >= ic->provided_data_sectors)) {
2477                                         journal_entry_set_unused(je);
2478                                         continue;
2479                                 }
2480                         }
2481                         get_area_and_offset(ic, sec, &area, &offset);
2482                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2483                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2484                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2485                                 sector_t sec2, area2, offset2;
2486                                 if (journal_entry_is_unused(je2))
2487                                         break;
2488                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2489                                 sec2 = journal_entry_get_sector(je2);
2490                                 if (unlikely(sec2 >= ic->provided_data_sectors))
2491                                         break;
2492                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2493                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2494                                         break;
2495                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2496                         }
2497                         next_loop = k - 1;
2498
2499                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2500                         io->comp = &comp;
2501                         io->range.logical_sector = sec;
2502                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2503
2504                         spin_lock_irq(&ic->endio_wait.lock);
2505                         add_new_range_and_wait(ic, &io->range);
2506
2507                         if (likely(!from_replay)) {
2508                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2509
2510                                 /* don't write if there is newer committed sector */
2511                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2512                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2513
2514                                         journal_entry_set_unused(je2);
2515                                         remove_journal_node(ic, &section_node[j]);
2516                                         j++;
2517                                         sec += ic->sectors_per_block;
2518                                         offset += ic->sectors_per_block;
2519                                 }
2520                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2521                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2522
2523                                         journal_entry_set_unused(je2);
2524                                         remove_journal_node(ic, &section_node[k - 1]);
2525                                         k--;
2526                                 }
2527                                 if (j == k) {
2528                                         remove_range_unlocked(ic, &io->range);
2529                                         spin_unlock_irq(&ic->endio_wait.lock);
2530                                         mempool_free(io, &ic->journal_io_mempool);
2531                                         goto skip_io;
2532                                 }
2533                                 for (l = j; l < k; l++) {
2534                                         remove_journal_node(ic, &section_node[l]);
2535                                 }
2536                         }
2537                         spin_unlock_irq(&ic->endio_wait.lock);
2538
2539                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2540                         for (l = j; l < k; l++) {
2541                                 int r;
2542                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2543
2544                                 if (
2545 #ifndef INTERNAL_VERIFY
2546                                     unlikely(from_replay) &&
2547 #endif
2548                                     ic->internal_hash) {
2549                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2550
2551                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2552                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2553                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size))) {
2554                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2555                                                 dm_audit_log_target(DM_MSG_PREFIX, "integrity-replay-journal", ic->ti, 0);
2556                                         }
2557                                 }
2558
2559                                 journal_entry_set_unused(je2);
2560                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2561                                                         ic->tag_size, TAG_WRITE);
2562                                 if (unlikely(r)) {
2563                                         dm_integrity_io_error(ic, "reading tags", r);
2564                                 }
2565                         }
2566
2567                         atomic_inc(&comp.in_flight);
2568                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2569                                           (k - j) << ic->sb->log2_sectors_per_block,
2570                                           get_data_sector(ic, area, offset),
2571                                           complete_copy_from_journal, io);
2572 skip_io:
2573                         j = next_loop;
2574                 }
2575         }
2576
2577         dm_bufio_write_dirty_buffers_async(ic->bufio);
2578
2579         blk_finish_plug(&plug);
2580
2581         complete_journal_op(&comp);
2582         wait_for_completion_io(&comp.comp);
2583
2584         dm_integrity_flush_buffers(ic, true);
2585 }
2586
2587 static void integrity_writer(struct work_struct *w)
2588 {
2589         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2590         unsigned write_start, write_sections;
2591
2592         unsigned prev_free_sectors;
2593
2594         /* the following test is not needed, but it tests the replay code */
2595         if (unlikely(dm_post_suspending(ic->ti)) && !ic->meta_dev)
2596                 return;
2597
2598         spin_lock_irq(&ic->endio_wait.lock);
2599         write_start = ic->committed_section;
2600         write_sections = ic->n_committed_sections;
2601         spin_unlock_irq(&ic->endio_wait.lock);
2602
2603         if (!write_sections)
2604                 return;
2605
2606         do_journal_write(ic, write_start, write_sections, false);
2607
2608         spin_lock_irq(&ic->endio_wait.lock);
2609
2610         ic->committed_section += write_sections;
2611         wraparound_section(ic, &ic->committed_section);
2612         ic->n_committed_sections -= write_sections;
2613
2614         prev_free_sectors = ic->free_sectors;
2615         ic->free_sectors += write_sections * ic->journal_section_entries;
2616         if (unlikely(!prev_free_sectors))
2617                 wake_up_locked(&ic->endio_wait);
2618
2619         spin_unlock_irq(&ic->endio_wait.lock);
2620 }
2621
2622 static void recalc_write_super(struct dm_integrity_c *ic)
2623 {
2624         int r;
2625
2626         dm_integrity_flush_buffers(ic, false);
2627         if (dm_integrity_failed(ic))
2628                 return;
2629
2630         r = sync_rw_sb(ic, REQ_OP_WRITE);
2631         if (unlikely(r))
2632                 dm_integrity_io_error(ic, "writing superblock", r);
2633 }
2634
2635 static void integrity_recalc(struct work_struct *w)
2636 {
2637         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2638         struct dm_integrity_range range;
2639         struct dm_io_request io_req;
2640         struct dm_io_region io_loc;
2641         sector_t area, offset;
2642         sector_t metadata_block;
2643         unsigned metadata_offset;
2644         sector_t logical_sector, n_sectors;
2645         __u8 *t;
2646         unsigned i;
2647         int r;
2648         unsigned super_counter = 0;
2649
2650         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2651
2652         spin_lock_irq(&ic->endio_wait.lock);
2653
2654 next_chunk:
2655
2656         if (unlikely(dm_post_suspending(ic->ti)))
2657                 goto unlock_ret;
2658
2659         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2660         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2661                 if (ic->mode == 'B') {
2662                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2663                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2664                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2665                 }
2666                 goto unlock_ret;
2667         }
2668
2669         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2670         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2671         if (!ic->meta_dev)
2672                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2673
2674         add_new_range_and_wait(ic, &range);
2675         spin_unlock_irq(&ic->endio_wait.lock);
2676         logical_sector = range.logical_sector;
2677         n_sectors = range.n_sectors;
2678
2679         if (ic->mode == 'B') {
2680                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2681                         goto advance_and_next;
2682                 }
2683                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2684                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2685                         logical_sector += ic->sectors_per_block;
2686                         n_sectors -= ic->sectors_per_block;
2687                         cond_resched();
2688                 }
2689                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2690                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2691                         n_sectors -= ic->sectors_per_block;
2692                         cond_resched();
2693                 }
2694                 get_area_and_offset(ic, logical_sector, &area, &offset);
2695         }
2696
2697         DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2698
2699         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2700                 recalc_write_super(ic);
2701                 if (ic->mode == 'B') {
2702                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2703                 }
2704                 super_counter = 0;
2705         }
2706
2707         if (unlikely(dm_integrity_failed(ic)))
2708                 goto err;
2709
2710         io_req.bi_opf = REQ_OP_READ;
2711         io_req.mem.type = DM_IO_VMA;
2712         io_req.mem.ptr.addr = ic->recalc_buffer;
2713         io_req.notify.fn = NULL;
2714         io_req.client = ic->io;
2715         io_loc.bdev = ic->dev->bdev;
2716         io_loc.sector = get_data_sector(ic, area, offset);
2717         io_loc.count = n_sectors;
2718
2719         r = dm_io(&io_req, 1, &io_loc, NULL);
2720         if (unlikely(r)) {
2721                 dm_integrity_io_error(ic, "reading data", r);
2722                 goto err;
2723         }
2724
2725         t = ic->recalc_tags;
2726         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2727                 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2728                 t += ic->tag_size;
2729         }
2730
2731         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2732
2733         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2734         if (unlikely(r)) {
2735                 dm_integrity_io_error(ic, "writing tags", r);
2736                 goto err;
2737         }
2738
2739         if (ic->mode == 'B') {
2740                 sector_t start, end;
2741                 start = (range.logical_sector >>
2742                          (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2743                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2744                 end = ((range.logical_sector + range.n_sectors) >>
2745                        (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2746                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2747                 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2748         }
2749
2750 advance_and_next:
2751         cond_resched();
2752
2753         spin_lock_irq(&ic->endio_wait.lock);
2754         remove_range_unlocked(ic, &range);
2755         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2756         goto next_chunk;
2757
2758 err:
2759         remove_range(ic, &range);
2760         return;
2761
2762 unlock_ret:
2763         spin_unlock_irq(&ic->endio_wait.lock);
2764
2765         recalc_write_super(ic);
2766 }
2767
2768 static void bitmap_block_work(struct work_struct *w)
2769 {
2770         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2771         struct dm_integrity_c *ic = bbs->ic;
2772         struct bio *bio;
2773         struct bio_list bio_queue;
2774         struct bio_list waiting;
2775
2776         bio_list_init(&waiting);
2777
2778         spin_lock(&bbs->bio_queue_lock);
2779         bio_queue = bbs->bio_queue;
2780         bio_list_init(&bbs->bio_queue);
2781         spin_unlock(&bbs->bio_queue_lock);
2782
2783         while ((bio = bio_list_pop(&bio_queue))) {
2784                 struct dm_integrity_io *dio;
2785
2786                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2787
2788                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2789                                     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2790                         remove_range(ic, &dio->range);
2791                         INIT_WORK(&dio->work, integrity_bio_wait);
2792                         queue_work(ic->offload_wq, &dio->work);
2793                 } else {
2794                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2795                                         dio->range.n_sectors, BITMAP_OP_SET);
2796                         bio_list_add(&waiting, bio);
2797                 }
2798         }
2799
2800         if (bio_list_empty(&waiting))
2801                 return;
2802
2803         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC,
2804                            bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2805                            BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2806
2807         while ((bio = bio_list_pop(&waiting))) {
2808                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2809
2810                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2811                                 dio->range.n_sectors, BITMAP_OP_SET);
2812
2813                 remove_range(ic, &dio->range);
2814                 INIT_WORK(&dio->work, integrity_bio_wait);
2815                 queue_work(ic->offload_wq, &dio->work);
2816         }
2817
2818         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2819 }
2820
2821 static void bitmap_flush_work(struct work_struct *work)
2822 {
2823         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2824         struct dm_integrity_range range;
2825         unsigned long limit;
2826         struct bio *bio;
2827
2828         dm_integrity_flush_buffers(ic, false);
2829
2830         range.logical_sector = 0;
2831         range.n_sectors = ic->provided_data_sectors;
2832
2833         spin_lock_irq(&ic->endio_wait.lock);
2834         add_new_range_and_wait(ic, &range);
2835         spin_unlock_irq(&ic->endio_wait.lock);
2836
2837         dm_integrity_flush_buffers(ic, true);
2838
2839         limit = ic->provided_data_sectors;
2840         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2841                 limit = le64_to_cpu(ic->sb->recalc_sector)
2842                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2843                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2844         }
2845         /*DEBUG_print("zeroing journal\n");*/
2846         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2847         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2848
2849         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
2850                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2851
2852         spin_lock_irq(&ic->endio_wait.lock);
2853         remove_range_unlocked(ic, &range);
2854         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2855                 bio_endio(bio);
2856                 spin_unlock_irq(&ic->endio_wait.lock);
2857                 spin_lock_irq(&ic->endio_wait.lock);
2858         }
2859         spin_unlock_irq(&ic->endio_wait.lock);
2860 }
2861
2862
2863 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2864                          unsigned n_sections, unsigned char commit_seq)
2865 {
2866         unsigned i, j, n;
2867
2868         if (!n_sections)
2869                 return;
2870
2871         for (n = 0; n < n_sections; n++) {
2872                 i = start_section + n;
2873                 wraparound_section(ic, &i);
2874                 for (j = 0; j < ic->journal_section_sectors; j++) {
2875                         struct journal_sector *js = access_journal(ic, i, j);
2876                         BUILD_BUG_ON(sizeof(js->sectors) != JOURNAL_SECTOR_DATA);
2877                         memset(&js->sectors, 0, sizeof(js->sectors));
2878                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2879                 }
2880                 for (j = 0; j < ic->journal_section_entries; j++) {
2881                         struct journal_entry *je = access_journal_entry(ic, i, j);
2882                         journal_entry_set_unused(je);
2883                 }
2884         }
2885
2886         write_journal(ic, start_section, n_sections);
2887 }
2888
2889 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2890 {
2891         unsigned char k;
2892         for (k = 0; k < N_COMMIT_IDS; k++) {
2893                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2894                         return k;
2895         }
2896         dm_integrity_io_error(ic, "journal commit id", -EIO);
2897         return -EIO;
2898 }
2899
2900 static void replay_journal(struct dm_integrity_c *ic)
2901 {
2902         unsigned i, j;
2903         bool used_commit_ids[N_COMMIT_IDS];
2904         unsigned max_commit_id_sections[N_COMMIT_IDS];
2905         unsigned write_start, write_sections;
2906         unsigned continue_section;
2907         bool journal_empty;
2908         unsigned char unused, last_used, want_commit_seq;
2909
2910         if (ic->mode == 'R')
2911                 return;
2912
2913         if (ic->journal_uptodate)
2914                 return;
2915
2916         last_used = 0;
2917         write_start = 0;
2918
2919         if (!ic->just_formatted) {
2920                 DEBUG_print("reading journal\n");
2921                 rw_journal(ic, REQ_OP_READ, 0, ic->journal_sections, NULL);
2922                 if (ic->journal_io)
2923                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2924                 if (ic->journal_io) {
2925                         struct journal_completion crypt_comp;
2926                         crypt_comp.ic = ic;
2927                         init_completion(&crypt_comp.comp);
2928                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2929                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2930                         wait_for_completion(&crypt_comp.comp);
2931                 }
2932                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2933         }
2934
2935         if (dm_integrity_failed(ic))
2936                 goto clear_journal;
2937
2938         journal_empty = true;
2939         memset(used_commit_ids, 0, sizeof used_commit_ids);
2940         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2941         for (i = 0; i < ic->journal_sections; i++) {
2942                 for (j = 0; j < ic->journal_section_sectors; j++) {
2943                         int k;
2944                         struct journal_sector *js = access_journal(ic, i, j);
2945                         k = find_commit_seq(ic, i, j, js->commit_id);
2946                         if (k < 0)
2947                                 goto clear_journal;
2948                         used_commit_ids[k] = true;
2949                         max_commit_id_sections[k] = i;
2950                 }
2951                 if (journal_empty) {
2952                         for (j = 0; j < ic->journal_section_entries; j++) {
2953                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2954                                 if (!journal_entry_is_unused(je)) {
2955                                         journal_empty = false;
2956                                         break;
2957                                 }
2958                         }
2959                 }
2960         }
2961
2962         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2963                 unused = N_COMMIT_IDS - 1;
2964                 while (unused && !used_commit_ids[unused - 1])
2965                         unused--;
2966         } else {
2967                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2968                         if (!used_commit_ids[unused])
2969                                 break;
2970                 if (unused == N_COMMIT_IDS) {
2971                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2972                         goto clear_journal;
2973                 }
2974         }
2975         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2976                     unused, used_commit_ids[0], used_commit_ids[1],
2977                     used_commit_ids[2], used_commit_ids[3]);
2978
2979         last_used = prev_commit_seq(unused);
2980         want_commit_seq = prev_commit_seq(last_used);
2981
2982         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2983                 journal_empty = true;
2984
2985         write_start = max_commit_id_sections[last_used] + 1;
2986         if (unlikely(write_start >= ic->journal_sections))
2987                 want_commit_seq = next_commit_seq(want_commit_seq);
2988         wraparound_section(ic, &write_start);
2989
2990         i = write_start;
2991         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2992                 for (j = 0; j < ic->journal_section_sectors; j++) {
2993                         struct journal_sector *js = access_journal(ic, i, j);
2994
2995                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2996                                 /*
2997                                  * This could be caused by crash during writing.
2998                                  * We won't replay the inconsistent part of the
2999                                  * journal.
3000                                  */
3001                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
3002                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
3003                                 goto brk;
3004                         }
3005                 }
3006                 i++;
3007                 if (unlikely(i >= ic->journal_sections))
3008                         want_commit_seq = next_commit_seq(want_commit_seq);
3009                 wraparound_section(ic, &i);
3010         }
3011 brk:
3012
3013         if (!journal_empty) {
3014                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
3015                             write_sections, write_start, want_commit_seq);
3016                 do_journal_write(ic, write_start, write_sections, true);
3017         }
3018
3019         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3020                 continue_section = write_start;
3021                 ic->commit_seq = want_commit_seq;
3022                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3023         } else {
3024                 unsigned s;
3025                 unsigned char erase_seq;
3026 clear_journal:
3027                 DEBUG_print("clearing journal\n");
3028
3029                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3030                 s = write_start;
3031                 init_journal(ic, s, 1, erase_seq);
3032                 s++;
3033                 wraparound_section(ic, &s);
3034                 if (ic->journal_sections >= 2) {
3035                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3036                         s += ic->journal_sections - 2;
3037                         wraparound_section(ic, &s);
3038                         init_journal(ic, s, 1, erase_seq);
3039                 }
3040
3041                 continue_section = 0;
3042                 ic->commit_seq = next_commit_seq(erase_seq);
3043         }
3044
3045         ic->committed_section = continue_section;
3046         ic->n_committed_sections = 0;
3047
3048         ic->uncommitted_section = continue_section;
3049         ic->n_uncommitted_sections = 0;
3050
3051         ic->free_section = continue_section;
3052         ic->free_section_entry = 0;
3053         ic->free_sectors = ic->journal_entries;
3054
3055         ic->journal_tree_root = RB_ROOT;
3056         for (i = 0; i < ic->journal_entries; i++)
3057                 init_journal_node(&ic->journal_tree[i]);
3058 }
3059
3060 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3061 {
3062         DEBUG_print("dm_integrity_enter_synchronous_mode\n");
3063
3064         if (ic->mode == 'B') {
3065                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3066                 ic->synchronous_mode = 1;
3067
3068                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3069                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3070                 flush_workqueue(ic->commit_wq);
3071         }
3072 }
3073
3074 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3075 {
3076         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3077
3078         DEBUG_print("dm_integrity_reboot\n");
3079
3080         dm_integrity_enter_synchronous_mode(ic);
3081
3082         return NOTIFY_DONE;
3083 }
3084
3085 static void dm_integrity_postsuspend(struct dm_target *ti)
3086 {
3087         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3088         int r;
3089
3090         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3091
3092         del_timer_sync(&ic->autocommit_timer);
3093
3094         if (ic->recalc_wq)
3095                 drain_workqueue(ic->recalc_wq);
3096
3097         if (ic->mode == 'B')
3098                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3099
3100         queue_work(ic->commit_wq, &ic->commit_work);
3101         drain_workqueue(ic->commit_wq);
3102
3103         if (ic->mode == 'J') {
3104                 if (ic->meta_dev)
3105                         queue_work(ic->writer_wq, &ic->writer_work);
3106                 drain_workqueue(ic->writer_wq);
3107                 dm_integrity_flush_buffers(ic, true);
3108         }
3109
3110         if (ic->mode == 'B') {
3111                 dm_integrity_flush_buffers(ic, true);
3112 #if 1
3113                 /* set to 0 to test bitmap replay code */
3114                 init_journal(ic, 0, ic->journal_sections, 0);
3115                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3116                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3117                 if (unlikely(r))
3118                         dm_integrity_io_error(ic, "writing superblock", r);
3119 #endif
3120         }
3121
3122         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3123
3124         ic->journal_uptodate = true;
3125 }
3126
3127 static void dm_integrity_resume(struct dm_target *ti)
3128 {
3129         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3130         __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3131         int r;
3132
3133         DEBUG_print("resume\n");
3134
3135         if (ic->provided_data_sectors != old_provided_data_sectors) {
3136                 if (ic->provided_data_sectors > old_provided_data_sectors &&
3137                     ic->mode == 'B' &&
3138                     ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3139                         rw_journal_sectors(ic, REQ_OP_READ, 0,
3140                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3141                         block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3142                                         ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3143                         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3144                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3145                 }
3146
3147                 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3148                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3149                 if (unlikely(r))
3150                         dm_integrity_io_error(ic, "writing superblock", r);
3151         }
3152
3153         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3154                 DEBUG_print("resume dirty_bitmap\n");
3155                 rw_journal_sectors(ic, REQ_OP_READ, 0,
3156                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3157                 if (ic->mode == 'B') {
3158                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3159                             !ic->reset_recalculate_flag) {
3160                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3161                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3162                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3163                                                      BITMAP_OP_TEST_ALL_CLEAR)) {
3164                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3165                                         ic->sb->recalc_sector = cpu_to_le64(0);
3166                                 }
3167                         } else {
3168                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3169                                             ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3170                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3171                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3172                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3173                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3174                                 rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3175                                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3176                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3177                                 ic->sb->recalc_sector = cpu_to_le64(0);
3178                         }
3179                 } else {
3180                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3181                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3182                             ic->reset_recalculate_flag) {
3183                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3184                                 ic->sb->recalc_sector = cpu_to_le64(0);
3185                         }
3186                         init_journal(ic, 0, ic->journal_sections, 0);
3187                         replay_journal(ic);
3188                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3189                 }
3190                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3191                 if (unlikely(r))
3192                         dm_integrity_io_error(ic, "writing superblock", r);
3193         } else {
3194                 replay_journal(ic);
3195                 if (ic->reset_recalculate_flag) {
3196                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3197                         ic->sb->recalc_sector = cpu_to_le64(0);
3198                 }
3199                 if (ic->mode == 'B') {
3200                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3201                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3202                         r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3203                         if (unlikely(r))
3204                                 dm_integrity_io_error(ic, "writing superblock", r);
3205
3206                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3207                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3208                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3209                         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3210                             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3211                                 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3212                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3213                                 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3214                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3215                                 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3216                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3217                         }
3218                         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3219                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3220                 }
3221         }
3222
3223         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3224         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3225                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3226                 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3227                 if (recalc_pos < ic->provided_data_sectors) {
3228                         queue_work(ic->recalc_wq, &ic->recalc_work);
3229                 } else if (recalc_pos > ic->provided_data_sectors) {
3230                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3231                         recalc_write_super(ic);
3232                 }
3233         }
3234
3235         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3236         ic->reboot_notifier.next = NULL;
3237         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
3238         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3239
3240 #if 0
3241         /* set to 1 to stress test synchronous mode */
3242         dm_integrity_enter_synchronous_mode(ic);
3243 #endif
3244 }
3245
3246 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3247                                 unsigned status_flags, char *result, unsigned maxlen)
3248 {
3249         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3250         unsigned arg_count;
3251         size_t sz = 0;
3252
3253         switch (type) {
3254         case STATUSTYPE_INFO:
3255                 DMEMIT("%llu %llu",
3256                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3257                         ic->provided_data_sectors);
3258                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3259                         DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3260                 else
3261                         DMEMIT(" -");
3262                 break;
3263
3264         case STATUSTYPE_TABLE: {
3265                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3266                 watermark_percentage += ic->journal_entries / 2;
3267                 do_div(watermark_percentage, ic->journal_entries);
3268                 arg_count = 3;
3269                 arg_count += !!ic->meta_dev;
3270                 arg_count += ic->sectors_per_block != 1;
3271                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3272                 arg_count += ic->reset_recalculate_flag;
3273                 arg_count += ic->discard;
3274                 arg_count += ic->mode == 'J';
3275                 arg_count += ic->mode == 'J';
3276                 arg_count += ic->mode == 'B';
3277                 arg_count += ic->mode == 'B';
3278                 arg_count += !!ic->internal_hash_alg.alg_string;
3279                 arg_count += !!ic->journal_crypt_alg.alg_string;
3280                 arg_count += !!ic->journal_mac_alg.alg_string;
3281                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3282                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3283                 arg_count += ic->legacy_recalculate;
3284                 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3285                        ic->tag_size, ic->mode, arg_count);
3286                 if (ic->meta_dev)
3287                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
3288                 if (ic->sectors_per_block != 1)
3289                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3290                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3291                         DMEMIT(" recalculate");
3292                 if (ic->reset_recalculate_flag)
3293                         DMEMIT(" reset_recalculate");
3294                 if (ic->discard)
3295                         DMEMIT(" allow_discards");
3296                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3297                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3298                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3299                 if (ic->mode == 'J') {
3300                         DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3301                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
3302                 }
3303                 if (ic->mode == 'B') {
3304                         DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3305                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3306                 }
3307                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3308                         DMEMIT(" fix_padding");
3309                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3310                         DMEMIT(" fix_hmac");
3311                 if (ic->legacy_recalculate)
3312                         DMEMIT(" legacy_recalculate");
3313
3314 #define EMIT_ALG(a, n)                                                  \
3315                 do {                                                    \
3316                         if (ic->a.alg_string) {                         \
3317                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
3318                                 if (ic->a.key_string)                   \
3319                                         DMEMIT(":%s", ic->a.key_string);\
3320                         }                                               \
3321                 } while (0)
3322                 EMIT_ALG(internal_hash_alg, "internal_hash");
3323                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3324                 EMIT_ALG(journal_mac_alg, "journal_mac");
3325                 break;
3326         }
3327         case STATUSTYPE_IMA:
3328                 DMEMIT_TARGET_NAME_VERSION(ti->type);
3329                 DMEMIT(",dev_name=%s,start=%llu,tag_size=%u,mode=%c",
3330                         ic->dev->name, ic->start, ic->tag_size, ic->mode);
3331
3332                 if (ic->meta_dev)
3333                         DMEMIT(",meta_device=%s", ic->meta_dev->name);
3334                 if (ic->sectors_per_block != 1)
3335                         DMEMIT(",block_size=%u", ic->sectors_per_block << SECTOR_SHIFT);
3336
3337                 DMEMIT(",recalculate=%c", (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) ?
3338                        'y' : 'n');
3339                 DMEMIT(",allow_discards=%c", ic->discard ? 'y' : 'n');
3340                 DMEMIT(",fix_padding=%c",
3341                        ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0) ? 'y' : 'n');
3342                 DMEMIT(",fix_hmac=%c",
3343                        ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0) ? 'y' : 'n');
3344                 DMEMIT(",legacy_recalculate=%c", ic->legacy_recalculate ? 'y' : 'n');
3345
3346                 DMEMIT(",journal_sectors=%u", ic->initial_sectors - SB_SECTORS);
3347                 DMEMIT(",interleave_sectors=%u", 1U << ic->sb->log2_interleave_sectors);
3348                 DMEMIT(",buffer_sectors=%u", 1U << ic->log2_buffer_sectors);
3349                 DMEMIT(";");
3350                 break;
3351         }
3352 }
3353
3354 static int dm_integrity_iterate_devices(struct dm_target *ti,
3355                                         iterate_devices_callout_fn fn, void *data)
3356 {
3357         struct dm_integrity_c *ic = ti->private;
3358
3359         if (!ic->meta_dev)
3360                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3361         else
3362                 return fn(ti, ic->dev, 0, ti->len, data);
3363 }
3364
3365 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3366 {
3367         struct dm_integrity_c *ic = ti->private;
3368
3369         if (ic->sectors_per_block > 1) {
3370                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3371                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3372                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3373         }
3374 }
3375
3376 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3377 {
3378         unsigned sector_space = JOURNAL_SECTOR_DATA;
3379
3380         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3381         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3382                                          JOURNAL_ENTRY_ROUNDUP);
3383
3384         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3385                 sector_space -= JOURNAL_MAC_PER_SECTOR;
3386         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3387         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3388         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3389         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3390 }
3391
3392 static int calculate_device_limits(struct dm_integrity_c *ic)
3393 {
3394         __u64 initial_sectors;
3395
3396         calculate_journal_section_size(ic);
3397         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3398         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3399                 return -EINVAL;
3400         ic->initial_sectors = initial_sectors;
3401
3402         if (!ic->meta_dev) {
3403                 sector_t last_sector, last_area, last_offset;
3404
3405                 /* we have to maintain excessive padding for compatibility with existing volumes */
3406                 __u64 metadata_run_padding =
3407                         ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3408                         (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3409                         (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3410
3411                 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3412                                             metadata_run_padding) >> SECTOR_SHIFT;
3413                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3414                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3415                 else
3416                         ic->log2_metadata_run = -1;
3417
3418                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3419                 last_sector = get_data_sector(ic, last_area, last_offset);
3420                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3421                         return -EINVAL;
3422         } else {
3423                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3424                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3425                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3426                 meta_size <<= ic->log2_buffer_sectors;
3427                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3428                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3429                         return -EINVAL;
3430                 ic->metadata_run = 1;
3431                 ic->log2_metadata_run = 0;
3432         }
3433
3434         return 0;
3435 }
3436
3437 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3438 {
3439         if (!ic->meta_dev) {
3440                 int test_bit;
3441                 ic->provided_data_sectors = 0;
3442                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3443                         __u64 prev_data_sectors = ic->provided_data_sectors;
3444
3445                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3446                         if (calculate_device_limits(ic))
3447                                 ic->provided_data_sectors = prev_data_sectors;
3448                 }
3449         } else {
3450                 ic->provided_data_sectors = ic->data_device_sectors;
3451                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3452         }
3453 }
3454
3455 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3456 {
3457         unsigned journal_sections;
3458         int test_bit;
3459
3460         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3461         memcpy(ic->sb->magic, SB_MAGIC, 8);
3462         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3463         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3464         if (ic->journal_mac_alg.alg_string)
3465                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3466
3467         calculate_journal_section_size(ic);
3468         journal_sections = journal_sectors / ic->journal_section_sectors;
3469         if (!journal_sections)
3470                 journal_sections = 1;
3471
3472         if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3473                 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3474                 get_random_bytes(ic->sb->salt, SALT_SIZE);
3475         }
3476
3477         if (!ic->meta_dev) {
3478                 if (ic->fix_padding)
3479                         ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3480                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3481                 if (!interleave_sectors)
3482                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3483                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3484                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3485                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3486
3487                 get_provided_data_sectors(ic);
3488                 if (!ic->provided_data_sectors)
3489                         return -EINVAL;
3490         } else {
3491                 ic->sb->log2_interleave_sectors = 0;
3492
3493                 get_provided_data_sectors(ic);
3494                 if (!ic->provided_data_sectors)
3495                         return -EINVAL;
3496
3497 try_smaller_buffer:
3498                 ic->sb->journal_sections = cpu_to_le32(0);
3499                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3500                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3501                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3502                         if (test_journal_sections > journal_sections)
3503                                 continue;
3504                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3505                         if (calculate_device_limits(ic))
3506                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3507
3508                 }
3509                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3510                         if (ic->log2_buffer_sectors > 3) {
3511                                 ic->log2_buffer_sectors--;
3512                                 goto try_smaller_buffer;
3513                         }
3514                         return -EINVAL;
3515                 }
3516         }
3517
3518         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3519
3520         sb_set_version(ic);
3521
3522         return 0;
3523 }
3524
3525 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3526 {
3527         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3528         struct blk_integrity bi;
3529
3530         memset(&bi, 0, sizeof(bi));
3531         bi.profile = &dm_integrity_profile;
3532         bi.tuple_size = ic->tag_size;
3533         bi.tag_size = bi.tuple_size;
3534         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3535
3536         blk_integrity_register(disk, &bi);
3537         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3538 }
3539
3540 static void dm_integrity_free_page_list(struct page_list *pl)
3541 {
3542         unsigned i;
3543
3544         if (!pl)
3545                 return;
3546         for (i = 0; pl[i].page; i++)
3547                 __free_page(pl[i].page);
3548         kvfree(pl);
3549 }
3550
3551 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3552 {
3553         struct page_list *pl;
3554         unsigned i;
3555
3556         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3557         if (!pl)
3558                 return NULL;
3559
3560         for (i = 0; i < n_pages; i++) {
3561                 pl[i].page = alloc_page(GFP_KERNEL);
3562                 if (!pl[i].page) {
3563                         dm_integrity_free_page_list(pl);
3564                         return NULL;
3565                 }
3566                 if (i)
3567                         pl[i - 1].next = &pl[i];
3568         }
3569         pl[i].page = NULL;
3570         pl[i].next = NULL;
3571
3572         return pl;
3573 }
3574
3575 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3576 {
3577         unsigned i;
3578         for (i = 0; i < ic->journal_sections; i++)
3579                 kvfree(sl[i]);
3580         kvfree(sl);
3581 }
3582
3583 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3584                                                                    struct page_list *pl)
3585 {
3586         struct scatterlist **sl;
3587         unsigned i;
3588
3589         sl = kvmalloc_array(ic->journal_sections,
3590                             sizeof(struct scatterlist *),
3591                             GFP_KERNEL | __GFP_ZERO);
3592         if (!sl)
3593                 return NULL;
3594
3595         for (i = 0; i < ic->journal_sections; i++) {
3596                 struct scatterlist *s;
3597                 unsigned start_index, start_offset;
3598                 unsigned end_index, end_offset;
3599                 unsigned n_pages;
3600                 unsigned idx;
3601
3602                 page_list_location(ic, i, 0, &start_index, &start_offset);
3603                 page_list_location(ic, i, ic->journal_section_sectors - 1,
3604                                    &end_index, &end_offset);
3605
3606                 n_pages = (end_index - start_index + 1);
3607
3608                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3609                                    GFP_KERNEL);
3610                 if (!s) {
3611                         dm_integrity_free_journal_scatterlist(ic, sl);
3612                         return NULL;
3613                 }
3614
3615                 sg_init_table(s, n_pages);
3616                 for (idx = start_index; idx <= end_index; idx++) {
3617                         char *va = lowmem_page_address(pl[idx].page);
3618                         unsigned start = 0, end = PAGE_SIZE;
3619                         if (idx == start_index)
3620                                 start = start_offset;
3621                         if (idx == end_index)
3622                                 end = end_offset + (1 << SECTOR_SHIFT);
3623                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3624                 }
3625
3626                 sl[i] = s;
3627         }
3628
3629         return sl;
3630 }
3631
3632 static void free_alg(struct alg_spec *a)
3633 {
3634         kfree_sensitive(a->alg_string);
3635         kfree_sensitive(a->key);
3636         memset(a, 0, sizeof *a);
3637 }
3638
3639 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3640 {
3641         char *k;
3642
3643         free_alg(a);
3644
3645         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3646         if (!a->alg_string)
3647                 goto nomem;
3648
3649         k = strchr(a->alg_string, ':');
3650         if (k) {
3651                 *k = 0;
3652                 a->key_string = k + 1;
3653                 if (strlen(a->key_string) & 1)
3654                         goto inval;
3655
3656                 a->key_size = strlen(a->key_string) / 2;
3657                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3658                 if (!a->key)
3659                         goto nomem;
3660                 if (hex2bin(a->key, a->key_string, a->key_size))
3661                         goto inval;
3662         }
3663
3664         return 0;
3665 inval:
3666         *error = error_inval;
3667         return -EINVAL;
3668 nomem:
3669         *error = "Out of memory for an argument";
3670         return -ENOMEM;
3671 }
3672
3673 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3674                    char *error_alg, char *error_key)
3675 {
3676         int r;
3677
3678         if (a->alg_string) {
3679                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3680                 if (IS_ERR(*hash)) {
3681                         *error = error_alg;
3682                         r = PTR_ERR(*hash);
3683                         *hash = NULL;
3684                         return r;
3685                 }
3686
3687                 if (a->key) {
3688                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3689                         if (r) {
3690                                 *error = error_key;
3691                                 return r;
3692                         }
3693                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3694                         *error = error_key;
3695                         return -ENOKEY;
3696                 }
3697         }
3698
3699         return 0;
3700 }
3701
3702 static int create_journal(struct dm_integrity_c *ic, char **error)
3703 {
3704         int r = 0;
3705         unsigned i;
3706         __u64 journal_pages, journal_desc_size, journal_tree_size;
3707         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3708         struct skcipher_request *req = NULL;
3709
3710         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3711         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3712         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3713         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3714
3715         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3716                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3717         journal_desc_size = journal_pages * sizeof(struct page_list);
3718         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3719                 *error = "Journal doesn't fit into memory";
3720                 r = -ENOMEM;
3721                 goto bad;
3722         }
3723         ic->journal_pages = journal_pages;
3724
3725         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3726         if (!ic->journal) {
3727                 *error = "Could not allocate memory for journal";
3728                 r = -ENOMEM;
3729                 goto bad;
3730         }
3731         if (ic->journal_crypt_alg.alg_string) {
3732                 unsigned ivsize, blocksize;
3733                 struct journal_completion comp;
3734
3735                 comp.ic = ic;
3736                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3737                 if (IS_ERR(ic->journal_crypt)) {
3738                         *error = "Invalid journal cipher";
3739                         r = PTR_ERR(ic->journal_crypt);
3740                         ic->journal_crypt = NULL;
3741                         goto bad;
3742                 }
3743                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3744                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3745
3746                 if (ic->journal_crypt_alg.key) {
3747                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3748                                                    ic->journal_crypt_alg.key_size);
3749                         if (r) {
3750                                 *error = "Error setting encryption key";
3751                                 goto bad;
3752                         }
3753                 }
3754                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3755                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3756
3757                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3758                 if (!ic->journal_io) {
3759                         *error = "Could not allocate memory for journal io";
3760                         r = -ENOMEM;
3761                         goto bad;
3762                 }
3763
3764                 if (blocksize == 1) {
3765                         struct scatterlist *sg;
3766
3767                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3768                         if (!req) {
3769                                 *error = "Could not allocate crypt request";
3770                                 r = -ENOMEM;
3771                                 goto bad;
3772                         }
3773
3774                         crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3775                         if (!crypt_iv) {
3776                                 *error = "Could not allocate iv";
3777                                 r = -ENOMEM;
3778                                 goto bad;
3779                         }
3780
3781                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3782                         if (!ic->journal_xor) {
3783                                 *error = "Could not allocate memory for journal xor";
3784                                 r = -ENOMEM;
3785                                 goto bad;
3786                         }
3787
3788                         sg = kvmalloc_array(ic->journal_pages + 1,
3789                                             sizeof(struct scatterlist),
3790                                             GFP_KERNEL);
3791                         if (!sg) {
3792                                 *error = "Unable to allocate sg list";
3793                                 r = -ENOMEM;
3794                                 goto bad;
3795                         }
3796                         sg_init_table(sg, ic->journal_pages + 1);
3797                         for (i = 0; i < ic->journal_pages; i++) {
3798                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3799                                 clear_page(va);
3800                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3801                         }
3802                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3803
3804                         skcipher_request_set_crypt(req, sg, sg,
3805                                                    PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3806                         init_completion(&comp.comp);
3807                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3808                         if (do_crypt(true, req, &comp))
3809                                 wait_for_completion(&comp.comp);
3810                         kvfree(sg);
3811                         r = dm_integrity_failed(ic);
3812                         if (r) {
3813                                 *error = "Unable to encrypt journal";
3814                                 goto bad;
3815                         }
3816                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3817
3818                         crypto_free_skcipher(ic->journal_crypt);
3819                         ic->journal_crypt = NULL;
3820                 } else {
3821                         unsigned crypt_len = roundup(ivsize, blocksize);
3822
3823                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3824                         if (!req) {
3825                                 *error = "Could not allocate crypt request";
3826                                 r = -ENOMEM;
3827                                 goto bad;
3828                         }
3829
3830                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3831                         if (!crypt_iv) {
3832                                 *error = "Could not allocate iv";
3833                                 r = -ENOMEM;
3834                                 goto bad;
3835                         }
3836
3837                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3838                         if (!crypt_data) {
3839                                 *error = "Unable to allocate crypt data";
3840                                 r = -ENOMEM;
3841                                 goto bad;
3842                         }
3843
3844                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3845                         if (!ic->journal_scatterlist) {
3846                                 *error = "Unable to allocate sg list";
3847                                 r = -ENOMEM;
3848                                 goto bad;
3849                         }
3850                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3851                         if (!ic->journal_io_scatterlist) {
3852                                 *error = "Unable to allocate sg list";
3853                                 r = -ENOMEM;
3854                                 goto bad;
3855                         }
3856                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3857                                                          sizeof(struct skcipher_request *),
3858                                                          GFP_KERNEL | __GFP_ZERO);
3859                         if (!ic->sk_requests) {
3860                                 *error = "Unable to allocate sk requests";
3861                                 r = -ENOMEM;
3862                                 goto bad;
3863                         }
3864                         for (i = 0; i < ic->journal_sections; i++) {
3865                                 struct scatterlist sg;
3866                                 struct skcipher_request *section_req;
3867                                 __le32 section_le = cpu_to_le32(i);
3868
3869                                 memset(crypt_iv, 0x00, ivsize);
3870                                 memset(crypt_data, 0x00, crypt_len);
3871                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3872
3873                                 sg_init_one(&sg, crypt_data, crypt_len);
3874                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3875                                 init_completion(&comp.comp);
3876                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3877                                 if (do_crypt(true, req, &comp))
3878                                         wait_for_completion(&comp.comp);
3879
3880                                 r = dm_integrity_failed(ic);
3881                                 if (r) {
3882                                         *error = "Unable to generate iv";
3883                                         goto bad;
3884                                 }
3885
3886                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3887                                 if (!section_req) {
3888                                         *error = "Unable to allocate crypt request";
3889                                         r = -ENOMEM;
3890                                         goto bad;
3891                                 }
3892                                 section_req->iv = kmalloc_array(ivsize, 2,
3893                                                                 GFP_KERNEL);
3894                                 if (!section_req->iv) {
3895                                         skcipher_request_free(section_req);
3896                                         *error = "Unable to allocate iv";
3897                                         r = -ENOMEM;
3898                                         goto bad;
3899                                 }
3900                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3901                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3902                                 ic->sk_requests[i] = section_req;
3903                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3904                         }
3905                 }
3906         }
3907
3908         for (i = 0; i < N_COMMIT_IDS; i++) {
3909                 unsigned j;
3910 retest_commit_id:
3911                 for (j = 0; j < i; j++) {
3912                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3913                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3914                                 goto retest_commit_id;
3915                         }
3916                 }
3917                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3918         }
3919
3920         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3921         if (journal_tree_size > ULONG_MAX) {
3922                 *error = "Journal doesn't fit into memory";
3923                 r = -ENOMEM;
3924                 goto bad;
3925         }
3926         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3927         if (!ic->journal_tree) {
3928                 *error = "Could not allocate memory for journal tree";
3929                 r = -ENOMEM;
3930         }
3931 bad:
3932         kfree(crypt_data);
3933         kfree(crypt_iv);
3934         skcipher_request_free(req);
3935
3936         return r;
3937 }
3938
3939 /*
3940  * Construct a integrity mapping
3941  *
3942  * Arguments:
3943  *      device
3944  *      offset from the start of the device
3945  *      tag size
3946  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3947  *      number of optional arguments
3948  *      optional arguments:
3949  *              journal_sectors
3950  *              interleave_sectors
3951  *              buffer_sectors
3952  *              journal_watermark
3953  *              commit_time
3954  *              meta_device
3955  *              block_size
3956  *              sectors_per_bit
3957  *              bitmap_flush_interval
3958  *              internal_hash
3959  *              journal_crypt
3960  *              journal_mac
3961  *              recalculate
3962  */
3963 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3964 {
3965         struct dm_integrity_c *ic;
3966         char dummy;
3967         int r;
3968         unsigned extra_args;
3969         struct dm_arg_set as;
3970         static const struct dm_arg _args[] = {
3971                 {0, 18, "Invalid number of feature args"},
3972         };
3973         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3974         bool should_write_sb;
3975         __u64 threshold;
3976         unsigned long long start;
3977         __s8 log2_sectors_per_bitmap_bit = -1;
3978         __s8 log2_blocks_per_bitmap_bit;
3979         __u64 bits_in_journal;
3980         __u64 n_bitmap_bits;
3981
3982 #define DIRECT_ARGUMENTS        4
3983
3984         if (argc <= DIRECT_ARGUMENTS) {
3985                 ti->error = "Invalid argument count";
3986                 return -EINVAL;
3987         }
3988
3989         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3990         if (!ic) {
3991                 ti->error = "Cannot allocate integrity context";
3992                 return -ENOMEM;
3993         }
3994         ti->private = ic;
3995         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3996         ic->ti = ti;
3997
3998         ic->in_progress = RB_ROOT;
3999         INIT_LIST_HEAD(&ic->wait_list);
4000         init_waitqueue_head(&ic->endio_wait);
4001         bio_list_init(&ic->flush_bio_list);
4002         init_waitqueue_head(&ic->copy_to_journal_wait);
4003         init_completion(&ic->crypto_backoff);
4004         atomic64_set(&ic->number_of_mismatches, 0);
4005         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
4006
4007         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
4008         if (r) {
4009                 ti->error = "Device lookup failed";
4010                 goto bad;
4011         }
4012
4013         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
4014                 ti->error = "Invalid starting offset";
4015                 r = -EINVAL;
4016                 goto bad;
4017         }
4018         ic->start = start;
4019
4020         if (strcmp(argv[2], "-")) {
4021                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
4022                         ti->error = "Invalid tag size";
4023                         r = -EINVAL;
4024                         goto bad;
4025                 }
4026         }
4027
4028         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
4029             !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
4030                 ic->mode = argv[3][0];
4031         } else {
4032                 ti->error = "Invalid mode (expecting J, B, D, R)";
4033                 r = -EINVAL;
4034                 goto bad;
4035         }
4036
4037         journal_sectors = 0;
4038         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4039         buffer_sectors = DEFAULT_BUFFER_SECTORS;
4040         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4041         sync_msec = DEFAULT_SYNC_MSEC;
4042         ic->sectors_per_block = 1;
4043
4044         as.argc = argc - DIRECT_ARGUMENTS;
4045         as.argv = argv + DIRECT_ARGUMENTS;
4046         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4047         if (r)
4048                 goto bad;
4049
4050         while (extra_args--) {
4051                 const char *opt_string;
4052                 unsigned val;
4053                 unsigned long long llval;
4054                 opt_string = dm_shift_arg(&as);
4055                 if (!opt_string) {
4056                         r = -EINVAL;
4057                         ti->error = "Not enough feature arguments";
4058                         goto bad;
4059                 }
4060                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4061                         journal_sectors = val ? val : 1;
4062                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4063                         interleave_sectors = val;
4064                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4065                         buffer_sectors = val;
4066                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4067                         journal_watermark = val;
4068                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4069                         sync_msec = val;
4070                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4071                         if (ic->meta_dev) {
4072                                 dm_put_device(ti, ic->meta_dev);
4073                                 ic->meta_dev = NULL;
4074                         }
4075                         r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4076                                           dm_table_get_mode(ti->table), &ic->meta_dev);
4077                         if (r) {
4078                                 ti->error = "Device lookup failed";
4079                                 goto bad;
4080                         }
4081                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4082                         if (val < 1 << SECTOR_SHIFT ||
4083                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4084                             (val & (val -1))) {
4085                                 r = -EINVAL;
4086                                 ti->error = "Invalid block_size argument";
4087                                 goto bad;
4088                         }
4089                         ic->sectors_per_block = val >> SECTOR_SHIFT;
4090                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4091                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4092                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4093                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4094                                 r = -EINVAL;
4095                                 ti->error = "Invalid bitmap_flush_interval argument";
4096                                 goto bad;
4097                         }
4098                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
4099                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4100                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4101                                             "Invalid internal_hash argument");
4102                         if (r)
4103                                 goto bad;
4104                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4105                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4106                                             "Invalid journal_crypt argument");
4107                         if (r)
4108                                 goto bad;
4109                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4110                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4111                                             "Invalid journal_mac argument");
4112                         if (r)
4113                                 goto bad;
4114                 } else if (!strcmp(opt_string, "recalculate")) {
4115                         ic->recalculate_flag = true;
4116                 } else if (!strcmp(opt_string, "reset_recalculate")) {
4117                         ic->recalculate_flag = true;
4118                         ic->reset_recalculate_flag = true;
4119                 } else if (!strcmp(opt_string, "allow_discards")) {
4120                         ic->discard = true;
4121                 } else if (!strcmp(opt_string, "fix_padding")) {
4122                         ic->fix_padding = true;
4123                 } else if (!strcmp(opt_string, "fix_hmac")) {
4124                         ic->fix_hmac = true;
4125                 } else if (!strcmp(opt_string, "legacy_recalculate")) {
4126                         ic->legacy_recalculate = true;
4127                 } else {
4128                         r = -EINVAL;
4129                         ti->error = "Invalid argument";
4130                         goto bad;
4131                 }
4132         }
4133
4134         ic->data_device_sectors = bdev_nr_sectors(ic->dev->bdev);
4135         if (!ic->meta_dev)
4136                 ic->meta_device_sectors = ic->data_device_sectors;
4137         else
4138                 ic->meta_device_sectors = bdev_nr_sectors(ic->meta_dev->bdev);
4139
4140         if (!journal_sectors) {
4141                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4142                                       ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4143         }
4144
4145         if (!buffer_sectors)
4146                 buffer_sectors = 1;
4147         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4148
4149         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4150                     "Invalid internal hash", "Error setting internal hash key");
4151         if (r)
4152                 goto bad;
4153
4154         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4155                     "Invalid journal mac", "Error setting journal mac key");
4156         if (r)
4157                 goto bad;
4158
4159         if (!ic->tag_size) {
4160                 if (!ic->internal_hash) {
4161                         ti->error = "Unknown tag size";
4162                         r = -EINVAL;
4163                         goto bad;
4164                 }
4165                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4166         }
4167         if (ic->tag_size > MAX_TAG_SIZE) {
4168                 ti->error = "Too big tag size";
4169                 r = -EINVAL;
4170                 goto bad;
4171         }
4172         if (!(ic->tag_size & (ic->tag_size - 1)))
4173                 ic->log2_tag_size = __ffs(ic->tag_size);
4174         else
4175                 ic->log2_tag_size = -1;
4176
4177         if (ic->mode == 'B' && !ic->internal_hash) {
4178                 r = -EINVAL;
4179                 ti->error = "Bitmap mode can be only used with internal hash";
4180                 goto bad;
4181         }
4182
4183         if (ic->discard && !ic->internal_hash) {
4184                 r = -EINVAL;
4185                 ti->error = "Discard can be only used with internal hash";
4186                 goto bad;
4187         }
4188
4189         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4190         ic->autocommit_msec = sync_msec;
4191         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4192
4193         ic->io = dm_io_client_create();
4194         if (IS_ERR(ic->io)) {
4195                 r = PTR_ERR(ic->io);
4196                 ic->io = NULL;
4197                 ti->error = "Cannot allocate dm io";
4198                 goto bad;
4199         }
4200
4201         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4202         if (r) {
4203                 ti->error = "Cannot allocate mempool";
4204                 goto bad;
4205         }
4206
4207         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4208                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4209         if (!ic->metadata_wq) {
4210                 ti->error = "Cannot allocate workqueue";
4211                 r = -ENOMEM;
4212                 goto bad;
4213         }
4214
4215         /*
4216          * If this workqueue were percpu, it would cause bio reordering
4217          * and reduced performance.
4218          */
4219         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4220         if (!ic->wait_wq) {
4221                 ti->error = "Cannot allocate workqueue";
4222                 r = -ENOMEM;
4223                 goto bad;
4224         }
4225
4226         ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4227                                           METADATA_WORKQUEUE_MAX_ACTIVE);
4228         if (!ic->offload_wq) {
4229                 ti->error = "Cannot allocate workqueue";
4230                 r = -ENOMEM;
4231                 goto bad;
4232         }
4233
4234         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4235         if (!ic->commit_wq) {
4236                 ti->error = "Cannot allocate workqueue";
4237                 r = -ENOMEM;
4238                 goto bad;
4239         }
4240         INIT_WORK(&ic->commit_work, integrity_commit);
4241
4242         if (ic->mode == 'J' || ic->mode == 'B') {
4243                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4244                 if (!ic->writer_wq) {
4245                         ti->error = "Cannot allocate workqueue";
4246                         r = -ENOMEM;
4247                         goto bad;
4248                 }
4249                 INIT_WORK(&ic->writer_work, integrity_writer);
4250         }
4251
4252         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4253         if (!ic->sb) {
4254                 r = -ENOMEM;
4255                 ti->error = "Cannot allocate superblock area";
4256                 goto bad;
4257         }
4258
4259         r = sync_rw_sb(ic, REQ_OP_READ);
4260         if (r) {
4261                 ti->error = "Error reading superblock";
4262                 goto bad;
4263         }
4264         should_write_sb = false;
4265         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4266                 if (ic->mode != 'R') {
4267                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4268                                 r = -EINVAL;
4269                                 ti->error = "The device is not initialized";
4270                                 goto bad;
4271                         }
4272                 }
4273
4274                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4275                 if (r) {
4276                         ti->error = "Could not initialize superblock";
4277                         goto bad;
4278                 }
4279                 if (ic->mode != 'R')
4280                         should_write_sb = true;
4281         }
4282
4283         if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4284                 r = -EINVAL;
4285                 ti->error = "Unknown version";
4286                 goto bad;
4287         }
4288         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4289                 r = -EINVAL;
4290                 ti->error = "Tag size doesn't match the information in superblock";
4291                 goto bad;
4292         }
4293         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4294                 r = -EINVAL;
4295                 ti->error = "Block size doesn't match the information in superblock";
4296                 goto bad;
4297         }
4298         if (!le32_to_cpu(ic->sb->journal_sections)) {
4299                 r = -EINVAL;
4300                 ti->error = "Corrupted superblock, journal_sections is 0";
4301                 goto bad;
4302         }
4303         /* make sure that ti->max_io_len doesn't overflow */
4304         if (!ic->meta_dev) {
4305                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4306                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4307                         r = -EINVAL;
4308                         ti->error = "Invalid interleave_sectors in the superblock";
4309                         goto bad;
4310                 }
4311         } else {
4312                 if (ic->sb->log2_interleave_sectors) {
4313                         r = -EINVAL;
4314                         ti->error = "Invalid interleave_sectors in the superblock";
4315                         goto bad;
4316                 }
4317         }
4318         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4319                 r = -EINVAL;
4320                 ti->error = "Journal mac mismatch";
4321                 goto bad;
4322         }
4323
4324         get_provided_data_sectors(ic);
4325         if (!ic->provided_data_sectors) {
4326                 r = -EINVAL;
4327                 ti->error = "The device is too small";
4328                 goto bad;
4329         }
4330
4331 try_smaller_buffer:
4332         r = calculate_device_limits(ic);
4333         if (r) {
4334                 if (ic->meta_dev) {
4335                         if (ic->log2_buffer_sectors > 3) {
4336                                 ic->log2_buffer_sectors--;
4337                                 goto try_smaller_buffer;
4338                         }
4339                 }
4340                 ti->error = "The device is too small";
4341                 goto bad;
4342         }
4343
4344         if (log2_sectors_per_bitmap_bit < 0)
4345                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4346         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4347                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4348
4349         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4350         if (bits_in_journal > UINT_MAX)
4351                 bits_in_journal = UINT_MAX;
4352         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4353                 log2_sectors_per_bitmap_bit++;
4354
4355         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4356         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4357         if (should_write_sb) {
4358                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4359         }
4360         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4361                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4362         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4363
4364         if (!ic->meta_dev)
4365                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4366
4367         if (ti->len > ic->provided_data_sectors) {
4368                 r = -EINVAL;
4369                 ti->error = "Not enough provided sectors for requested mapping size";
4370                 goto bad;
4371         }
4372
4373
4374         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4375         threshold += 50;
4376         do_div(threshold, 100);
4377         ic->free_sectors_threshold = threshold;
4378
4379         DEBUG_print("initialized:\n");
4380         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4381         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
4382         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4383         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
4384         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
4385         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4386         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
4387         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4388         DEBUG_print("   data_device_sectors 0x%llx\n", bdev_nr_sectors(ic->dev->bdev));
4389         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
4390         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
4391         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
4392         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4393         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4394         DEBUG_print("   bits_in_journal %llu\n", bits_in_journal);
4395
4396         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4397                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4398                 ic->sb->recalc_sector = cpu_to_le64(0);
4399         }
4400
4401         if (ic->internal_hash) {
4402                 size_t recalc_tags_size;
4403                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4404                 if (!ic->recalc_wq ) {
4405                         ti->error = "Cannot allocate workqueue";
4406                         r = -ENOMEM;
4407                         goto bad;
4408                 }
4409                 INIT_WORK(&ic->recalc_work, integrity_recalc);
4410                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4411                 if (!ic->recalc_buffer) {
4412                         ti->error = "Cannot allocate buffer for recalculating";
4413                         r = -ENOMEM;
4414                         goto bad;
4415                 }
4416                 recalc_tags_size = (RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size;
4417                 if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
4418                         recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
4419                 ic->recalc_tags = kvmalloc(recalc_tags_size, GFP_KERNEL);
4420                 if (!ic->recalc_tags) {
4421                         ti->error = "Cannot allocate tags for recalculating";
4422                         r = -ENOMEM;
4423                         goto bad;
4424                 }
4425         } else {
4426                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4427                         ti->error = "Recalculate can only be specified with internal_hash";
4428                         r = -EINVAL;
4429                         goto bad;
4430                 }
4431         }
4432
4433         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4434             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4435             dm_integrity_disable_recalculate(ic)) {
4436                 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4437                 r = -EOPNOTSUPP;
4438                 goto bad;
4439         }
4440
4441         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4442                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL, 0);
4443         if (IS_ERR(ic->bufio)) {
4444                 r = PTR_ERR(ic->bufio);
4445                 ti->error = "Cannot initialize dm-bufio";
4446                 ic->bufio = NULL;
4447                 goto bad;
4448         }
4449         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4450
4451         if (ic->mode != 'R') {
4452                 r = create_journal(ic, &ti->error);
4453                 if (r)
4454                         goto bad;
4455
4456         }
4457
4458         if (ic->mode == 'B') {
4459                 unsigned i;
4460                 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4461
4462                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4463                 if (!ic->recalc_bitmap) {
4464                         r = -ENOMEM;
4465                         goto bad;
4466                 }
4467                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4468                 if (!ic->may_write_bitmap) {
4469                         r = -ENOMEM;
4470                         goto bad;
4471                 }
4472                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4473                 if (!ic->bbs) {
4474                         r = -ENOMEM;
4475                         goto bad;
4476                 }
4477                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4478                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4479                         struct bitmap_block_status *bbs = &ic->bbs[i];
4480                         unsigned sector, pl_index, pl_offset;
4481
4482                         INIT_WORK(&bbs->work, bitmap_block_work);
4483                         bbs->ic = ic;
4484                         bbs->idx = i;
4485                         bio_list_init(&bbs->bio_queue);
4486                         spin_lock_init(&bbs->bio_queue_lock);
4487
4488                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4489                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4490                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4491
4492                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4493                 }
4494         }
4495
4496         if (should_write_sb) {
4497                 init_journal(ic, 0, ic->journal_sections, 0);
4498                 r = dm_integrity_failed(ic);
4499                 if (unlikely(r)) {
4500                         ti->error = "Error initializing journal";
4501                         goto bad;
4502                 }
4503                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
4504                 if (r) {
4505                         ti->error = "Error initializing superblock";
4506                         goto bad;
4507                 }
4508                 ic->just_formatted = true;
4509         }
4510
4511         if (!ic->meta_dev) {
4512                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4513                 if (r)
4514                         goto bad;
4515         }
4516         if (ic->mode == 'B') {
4517                 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4518                 if (!max_io_len)
4519                         max_io_len = 1U << 31;
4520                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4521                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4522                         r = dm_set_target_max_io_len(ti, max_io_len);
4523                         if (r)
4524                                 goto bad;
4525                 }
4526         }
4527
4528         if (!ic->internal_hash)
4529                 dm_integrity_set(ti, ic);
4530
4531         ti->num_flush_bios = 1;
4532         ti->flush_supported = true;
4533         if (ic->discard)
4534                 ti->num_discard_bios = 1;
4535
4536         dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
4537         return 0;
4538
4539 bad:
4540         dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
4541         dm_integrity_dtr(ti);
4542         return r;
4543 }
4544
4545 static void dm_integrity_dtr(struct dm_target *ti)
4546 {
4547         struct dm_integrity_c *ic = ti->private;
4548
4549         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4550         BUG_ON(!list_empty(&ic->wait_list));
4551
4552         if (ic->metadata_wq)
4553                 destroy_workqueue(ic->metadata_wq);
4554         if (ic->wait_wq)
4555                 destroy_workqueue(ic->wait_wq);
4556         if (ic->offload_wq)
4557                 destroy_workqueue(ic->offload_wq);
4558         if (ic->commit_wq)
4559                 destroy_workqueue(ic->commit_wq);
4560         if (ic->writer_wq)
4561                 destroy_workqueue(ic->writer_wq);
4562         if (ic->recalc_wq)
4563                 destroy_workqueue(ic->recalc_wq);
4564         vfree(ic->recalc_buffer);
4565         kvfree(ic->recalc_tags);
4566         kvfree(ic->bbs);
4567         if (ic->bufio)
4568                 dm_bufio_client_destroy(ic->bufio);
4569         mempool_exit(&ic->journal_io_mempool);
4570         if (ic->io)
4571                 dm_io_client_destroy(ic->io);
4572         if (ic->dev)
4573                 dm_put_device(ti, ic->dev);
4574         if (ic->meta_dev)
4575                 dm_put_device(ti, ic->meta_dev);
4576         dm_integrity_free_page_list(ic->journal);
4577         dm_integrity_free_page_list(ic->journal_io);
4578         dm_integrity_free_page_list(ic->journal_xor);
4579         dm_integrity_free_page_list(ic->recalc_bitmap);
4580         dm_integrity_free_page_list(ic->may_write_bitmap);
4581         if (ic->journal_scatterlist)
4582                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4583         if (ic->journal_io_scatterlist)
4584                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4585         if (ic->sk_requests) {
4586                 unsigned i;
4587
4588                 for (i = 0; i < ic->journal_sections; i++) {
4589                         struct skcipher_request *req = ic->sk_requests[i];
4590                         if (req) {
4591                                 kfree_sensitive(req->iv);
4592                                 skcipher_request_free(req);
4593                         }
4594                 }
4595                 kvfree(ic->sk_requests);
4596         }
4597         kvfree(ic->journal_tree);
4598         if (ic->sb)
4599                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4600
4601         if (ic->internal_hash)
4602                 crypto_free_shash(ic->internal_hash);
4603         free_alg(&ic->internal_hash_alg);
4604
4605         if (ic->journal_crypt)
4606                 crypto_free_skcipher(ic->journal_crypt);
4607         free_alg(&ic->journal_crypt_alg);
4608
4609         if (ic->journal_mac)
4610                 crypto_free_shash(ic->journal_mac);
4611         free_alg(&ic->journal_mac_alg);
4612
4613         kfree(ic);
4614         dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
4615 }
4616
4617 static struct target_type integrity_target = {
4618         .name                   = "integrity",
4619         .version                = {1, 10, 0},
4620         .module                 = THIS_MODULE,
4621         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4622         .ctr                    = dm_integrity_ctr,
4623         .dtr                    = dm_integrity_dtr,
4624         .map                    = dm_integrity_map,
4625         .postsuspend            = dm_integrity_postsuspend,
4626         .resume                 = dm_integrity_resume,
4627         .status                 = dm_integrity_status,
4628         .iterate_devices        = dm_integrity_iterate_devices,
4629         .io_hints               = dm_integrity_io_hints,
4630 };
4631
4632 static int __init dm_integrity_init(void)
4633 {
4634         int r;
4635
4636         journal_io_cache = kmem_cache_create("integrity_journal_io",
4637                                              sizeof(struct journal_io), 0, 0, NULL);
4638         if (!journal_io_cache) {
4639                 DMERR("can't allocate journal io cache");
4640                 return -ENOMEM;
4641         }
4642
4643         r = dm_register_target(&integrity_target);
4644
4645         if (r < 0)
4646                 DMERR("register failed %d", r);
4647
4648         return r;
4649 }
4650
4651 static void __exit dm_integrity_exit(void)
4652 {
4653         dm_unregister_target(&integrity_target);
4654         kmem_cache_destroy(journal_io_cache);
4655 }
4656
4657 module_init(dm_integrity_init);
4658 module_exit(dm_integrity_exit);
4659
4660 MODULE_AUTHOR("Milan Broz");
4661 MODULE_AUTHOR("Mikulas Patocka");
4662 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4663 MODULE_LICENSE("GPL");